# roadkeep changelog — alegauss

Every entry in roadkeep's roadkeep ledger, dated by the commit that wrote it.

- 2026-10-18 · RK1713: no writing verb can be asked what it would write, so an agent probes a replace with a real write
- 2026-10-18 · RK1712: show refuses a deferred id and only names where it sits, so a paused line's section is read by grep
- 2026-10-18 · RK1711: section add --role strategy refuses every anchor on a fresh init --strategy scaffold, and its remedy is the refused call
- 2026-10-18 · RK1710: non-goal list elides entries past twelve, and neither --json nor any flag prints the rest
- 2026-10-18 · RK1709: anchors --next names no free top-level when a family is not one numbering
- 2026-09-28 · RK1708: a ledger entry whose id the grammar rejects has no verb that corrects it, and its finding names one that refuses
- 2026-09-28 · RK1707: code.renamed fires in an adopter that vendors the engine, so its gate reports renames that never happened
- 2026-09-28 · RK1706: a line's requires can be replaced but never cleared, so a need a person met stays on the line
- 2026-09-28 · RK1705: a launcher committed before RK1699 finds no engine in a roadkeep checkout, and its guard then stands down unsaid
- 2026-09-25 · RK1704: The reader's public site builds here but deploys nowhere, since this repository's one Pages site is roadkeep's
- 2026-09-25 · RK1702: A release tag here publishes the CLI and builds no installer, so the app ships from a different commit
- 2026-09-25 · RK1701: The installer cannot tell a machine without Python 3.11 or the claude CLI from one where the plugin installed
- 2026-09-25 · RK1700: Installing the reader and the Claude Code plugin takes two separate procedures on each of three platforms
- 2026-09-25 · RK1699: The plugin's source is the repository root, so installing it would copy the Electron app into every adopter's cache
- 2026-09-24 · RK1703: A session under gui/ loads a dev skill that knows pytest and run-commit.cmd, not npm gates or staging by path
- 2026-09-24 · RK1698: With gui/ in the tree, lint, the guard hook and MCP started at the root govern RK only, so gui/docs goes unguarded
- 2026-09-24 · RK1696: The non-goal against a web UI does not say whether a desktop reader that only spawns the CLI may live here
- 2026-09-21 · RK1695: a record amend whose tail the parser would not read back crashes with a traceback instead of refusing
- 2026-09-21 · RK1694: a validation that failed is a defect with nowhere to go, so the verdict and its line are two commands
- 2026-09-21 · RK1693: a verdict line is invisible to `carries`, so an entry holding one reads as hand-wrapped to `record amend`
- 2026-09-21 · RK1692: a project adopting validation opens with every entry it ever shipped in the list, 1163 of them here
- 2026-09-21 · RK1691: nothing answers which shipped entries no person has looked at, so the list is read by eye
- 2026-09-21 · RK1690: the ledger says what shipped and nothing says whether a person ever tried it
- 2026-09-21 · RK1689: The guard asks about every commit that stages a governed file, though git add and git commit never write it
- 2026-09-16 · RK1688: `show` drops a path written as a bare filename, so a section naming five sprites reports none of them
- 2026-09-16 · RK1687: `evidence` matches a regex over text, so a task whose proof is a number out of a PNG or a WAV declares none
- 2026-09-16 · RK1686: `weight` reads a commit whole, so a task that regenerated 37 sprites prices at 770 lines against 78 authored
- 2026-09-16 · RK1685: show returns one section, so reading the designs behind a set of related lines is one call each
- 2026-09-12 · RK1684: `[files]` takes a path that climbs out of the project root, and its absolute-path refusal misses `/x` on Windows
- 2026-09-12 · RK1683: an emoji is all `[project]` can point at, so a repository with a real logo has nowhere to name the file
- 2026-09-12 · RK1682: a reader of many checkouts has only the folder name: roadkeep.toml declares no name, description or icon
- 2026-09-12 · RK1681: a refusal on the decisions role hands back a budget door that measures the draft against the improvements limit
- 2026-09-12 · RK1680: list prints no readiness per line, so a consumer resolves deps once per row to draw a column it already counted
- 2026-09-12 · RK1679: a source line uses 3.12 f-string syntax while the manifest promises 3.11, so the package does not import there
- 2026-09-11 · RK1678: the launcher drops an engine the caller named when its probe fails, and answers from a clone hundreds of versions old
- 2026-09-11 · RK1677: list --role deferred --json carries no age for a pause, which --stale computes and prints only for a terminal
- 2026-09-10 · RK1676: non-goal list --json publishes each non-goal's lead and not its reason, which criterion list carries as why
- 2026-09-10 · RK1675: The door file.missing names is init, which refuses on the configured project that emits the finding
- 2026-09-10 · RK1674: Two more refusals over an undeclared role name the vocabulary and not the verb that adds one
- 2026-09-10 · RK1673: No verb writes the two opt-in tables' numbers, and two messages say govern does
- 2026-09-10 · RK1672: the retry a refusal hands back carries a token no shell delivers, one line above the door that does not
- 2026-09-10 · RK1671: three tests assert on this checkout's live backlog and go red when it drains
- 2026-09-10 · RK1670: the refusal over an undeclared role names what the project has and not the verb that adds one
- 2026-09-10 · RK1669: a subjectless served call is answered with rows naming subjects that transport withholds
- 2026-09-10 · RK1668: thirteen printed doors name a command with no invocation, so a caller pastes a line their shell has no such command for
- 2026-09-10 · RK1667: the report door mangles a symptom carrying a backtick in the shell this repository is developed in
- 2026-09-09 · RK1666: govern writes a caller's sentence into roadkeep.toml and no validator reads it, so a mangled run lands and stays
- 2026-09-09 · RK1665: the backstop reads a code only where it is a literal, so six a write refuses are outside a total that says it is one
- 2026-09-09 · RK1663: the per-block half of a pin's shape counts the roadmap's headings, so a finished block's entries land in no row
- 2026-09-09 · RK1656: a dep on a ruled-out finding resolves as unknown, which says nothing can answer about a file that answers exactly
- 2026-09-09 · RK1655: reopen files a line whose pointer resolves to nothing, and has no way to write the design in the same call
- 2026-09-09 · RK1654: ten functions print to both streams in one body and two go through the helper, nothing saying which of the rest need to
- 2026-09-09 · RK1653: five classes answer to notes and one is a gate report, so a sweep over the name is off for three whole test modules
- 2026-09-09 · RK1652: a config refusal now names the table a key belongs under, and no verb moves it there
- 2026-09-09 · RK1651: three readings of one source claim to be complements of each other and no test holds any of the three claims
- 2026-09-09 · RK1650: one of forty read-only verbs now refuses, and the rule that reading is never refused is prose in three places
- 2026-09-09 · RK1649: answers() spells at most one of a verb's flags and nothing spells exactly one, so a required choice stays a raise
- 2026-09-09 · RK1648: verifying a vendored engine runs it, so Python writes three megabytes of bytecode into an artefact just measured at four
- 2026-09-09 · RK1647: three walkers in the suite each rebuild which function a node sits in, and one of them spells the address differently
- 2026-09-09 · RK1646: the guard against the folded idiom sweeps the package and not the suite, where two sites still spell it out
- 2026-09-09 · RK1645: config publishes three lists of objects and the promise table holds one slot per verb, so one list's rows are unpromised
- 2026-09-09 · RK1644: nine guards over this package's source read its characters where twenty-two read its syntax
- 2026-09-09 · RK1643: the orientation is held at a ceiling and the two pages it points at, six times its size, are held by nothing
- 2026-09-09 · RK1642: the preventive read a refusal names for seven codes is in the sentence and in no field of the payload
- 2026-09-09 · RK1641: the sweep that runs every door the gate offers reaches one of eighty-four dispatchable rows
- 2026-09-09 · RK1640: 421 verb-leading spans sit in messages carrying no sibling door, so the pair that decides one reads none of them
- 2026-09-09 · RK1639: one closed set of prose was swept for stale figures and the three a caller actually meets were not
- 2026-09-09 · RK1638: a prose check finds its span by splitting on a sentence, so a reworded clause empties the test instead of failing it
- 2026-09-09 · RK1637: seven cost subjects each argued they were the Nth cadence nobody counted, and nothing enumerates the cadences
- 2026-09-09 · RK1636: a name a module imports and then defines is bound by the later one, and nothing here reports the dead import
- 2026-09-09 · RK1635: the composer sweep runs every door through shlex, which is the assumption the door was wrong about
- 2026-09-09 · RK1634: a ship deletes the design section and names nothing that was in it, so the last look at it never happens
- 2026-09-09 · RK1633: one verb handler writes a governed file itself, and it is the writer every enumeration of the config's has missed
- 2026-09-09 · RK1632: add --symptom takes no stdin, the one prose field with no pipe being the one every line leads with
- 2026-09-09 · RK1631: no one call says whether a path is governed and where its root is, so a client reads it off a null
- 2026-09-09 · RK1630: a read payload names neither the root it answered about nor the build that answered
- 2026-09-09 · RK1629: two modules walk back from a file's last blank line under two names, and the kernel that owns blank has no name for it
- 2026-09-09 · RK1628: pick offers an idea whose premise a later ship deleted, no answer saying what landed under the line since it was filed
- 2026-09-09 · RK1627: nothing enumerates the fields a caller composes, so which of them has a validator is answered by remembering
- 2026-09-09 · RK1626: the field sweep skips its whole assertion where a corpus is absent, including this repository's own 2,246 fields
- 2026-09-08 · RK1625: unclosed answers a tree with no git as a backlog with nothing open, the flag that tells those apart never being set
- 2026-09-08 · RK1624: the ranking an add volunteers exists only inside that write, no read answering over a block's whole corpus
- 2026-09-08 · RK1623: three tests rebuild the corpus add ranks against, so a change to what add includes leaves every figure passing
- 2026-09-08 · RK1622: a design can cite a corpus for a ratio it does not have, nothing here reading what shape the pins hold
- 2026-09-08 · RK1621: the summary and the folded note row now both carry the count of surfaces behind the engine
- 2026-09-08 · RK1620: the note census prices one row per note, so it reports what the gate composed and not what it printed
- 2026-09-07 · RK1619: the note sweep names the first call in an answer, so a helper before the note would take its kind
- 2026-09-07 · RK1618: delivered and reversals are the reads before an add, and neither holds what was traced and not filed
- 2026-09-07 · RK1612: a verb writing an answer and a note in one run has to flush between the streams, and nothing says which verbs those are
- 2026-09-07 · RK1611: uninstall keeps a server declaration it did not write and takes away the approval that made it run
- 2026-09-07 · RK1610: a config key typed under the wrong table is refused as unknown, where the tool knows exactly which table it belongs to
- 2026-09-07 · RK1609: three scans of one rule reconstruct a composed command three ways, and a value sits in the gap between them
- 2026-09-07 · RK1608: a narrowing flag whose vocabulary the project declared nothing for filters nothing and says nothing
- 2026-09-07 · RK1607: four two-answer rules are raised inside handlers where the declaration that spells them is one call away
- 2026-09-07 · RK1606: a vendored engine is twenty-two megabytes of which under four are the engine, the rest a built site and a test suite
- 2026-09-07 · RK1605: the composer census counts sites that call the invocation, so a door spelled without it is outside the guarantee
- 2026-09-07 · RK1604: the one reader saying whether a role is available answers a third question and has no docstring
- 2026-09-07 · RK1603: a table's note is repeated on every key under it, and that repetition is 29 percent of the config payload
- 2026-09-07 · RK1602: the decisions pointer says it is built from the constant that writes it and spells the literal again
- 2026-09-07 · RK1601: fifty-eight served arguments across thirty-three tools are spelled in no call the guidance shows
- 2026-09-07 · RK1600: a refusal payload publishes rules and never the retry the same refusal composed
- 2026-09-07 · RK1591: a remedy whose first step is an edit outside this tool has no kind, so repair dispatches a door that refuses
- 2026-09-07 · RK1590: ninety backticked spans outside a docstring name a verb with no invocation, so nothing can tell a door from prose
- 2026-09-07 · RK1588: a withholding reason quoted a total that was stale within the session that read it, and no sweep says how many are
- 2026-09-07 · RK1585: prose beside a total table is taken for the table, and one check reads one phrase in one comment
- 2026-09-07 · RK1583: a deferral says nothing about the limit its reason was measured against, only govern does
- 2026-09-07 · RK1582: the rows an add volunteers are prose on every write and no cadence counts them
- 2026-09-07 · RK1581: install reads roadkeep.toml beside the tree where every other verb discovers it, so a governed subtree reads as none
- 2026-09-07 · RK1580: the capture's filing door is POSIX-quoted on a platform whose shell is not, and nothing has run it there
- 2026-09-07 · RK1576: the config read-back is one writer's, and the decision RK1533 filed binds five
- 2026-09-07 · RK1573: declare non_goals opens a table whose verb refuses, nothing writing the heading that declares the list
- 2026-09-06 · RK1617: a handler answers in a printed register and an exit code, so no second surface can take its answer as a value
- 2026-09-06 · RK1616: nothing refuses a handler whose answer is not a result, so the shape is held by review alone
- 2026-09-06 · RK1615: the served surface reads a handler's answer by capturing the stdout it printed for a terminal
- 2026-09-06 · RK1614: ninety-five payload builders answer in twenty signatures and nothing declares the shape they share
- 2026-09-06 · RK1613: a refusal reaches an agent as prose where the same refusal reaches a terminal as fields
- 2026-09-06 · RK1599: two sites remain a work-list, and the door a capture names cannot be read back at all
- 2026-09-06 · RK1598: three sites that compose a read are a work-list, and each of the three sentences says something untrue
- 2026-09-06 · RK1597: two sites that compose a read about a line nothing counted are a work-list, and one is quoted with apostrophes
- 2026-09-06 · RK1596: three sites that compose a read about the ledger are a work-list, and one of them names two doors
- 2026-09-06 · RK1595: three of the sites that compose a read are accounted for as a work-list, and one was unrunnable by the instrument
- 2026-09-06 · RK1594: two of the sites that compose an outline read are accounted for as a work-list, and one of them cannot be reached at all
- 2026-09-06 · RK1593: two of the sites that compose a door out of a pause are accounted for as a work-list, so nothing runs one
- 2026-09-06 · RK1592: three of the sites that compose a gate door are accounted for as a work-list, so nothing runs one
- 2026-09-06 · RK1589: the merge row an install prints names three doors and nothing has ever run one of them
- 2026-09-06 · RK1587: ten records of six or more fields are built positionally, and which of them more than one module builds is unmeasured
- 2026-09-06 · RK1586: a shared record is built positionally by one of its callers, so a field inserted above another moves a figure
- 2026-09-06 · RK1584: a refused call publishes no payload, so every field of a violation reaches an agent as prose to match on
- 2026-09-06 · RK1579: two more sites of the composer work-list are accounted for, one of them as work it never was
- 2026-09-06 · RK1577: two more sites of the composer work-list compose doors nothing has ever run
- 2026-09-06 · RK1575: three more sites of the composer work-list compose doors nothing has ever run
- 2026-09-06 · RK1572: three sites of the composer work-list compose doors nothing has ever run
- 2026-09-06 · RK1570: a block title is written into every governed file and nothing validates it, not even the codec rule
- 2026-09-06 · RK1569: the prose half of the re-taken split reds on somebody else's edit, which says nothing about the rule
- 2026-09-06 · RK1568: an incidental entry that is on disk and filters no commit is silent, which is the half of the reading RK1529 gave
- 2026-09-06 · RK1567: the near rows are described as delivered --near volunteered, and that read has ranked a different corpus since RK1495
- 2026-09-06 · RK1566: the near window is measured on a ledger far larger than its backlog, the ratio it is least about
- 2026-09-06 · RK1565: the per-surface staleness rows may repeat one door six times, and nothing has measured what that costs
- 2026-09-06 · RK1564: the note sweep reads every call inside an answer's text, so a helper beside a note reads as a kind
- 2026-09-06 · RK1563: a field that shares a name with the served prefix is reported as a carrier of it, which it is not
- 2026-09-06 · RK1562: the served notes are priced against a module list built here rather than the one a real note carries
- 2026-09-06 · RK1561: engines --invoke prints a command and never that the declaration it fell through was one this tool cannot read
- 2026-09-06 · RK1560: install replaces a server declaration an adopter wrote themselves and no row names what was there
- 2026-09-06 · RK1559: twelve of the fifteen note codes have never been priced, this project's gate tripping only two of them
- 2026-09-06 · RK1558: one marker scan skips what a caller is shown and the one beside it does not, with nothing saying they differ
- 2026-09-06 · RK1557: the gate against a composed marker reads a bare name, so the same constant reached through its module walks past
- 2026-09-06 · RK1556: a project whose open set spells no partial marker ships a partial in silence, and the line keeps the marker it had
- 2026-09-06 · RK1555: two of adopt's argument rules are still raised inside the estimator, where no declaration reader can see them
- 2026-09-06 · RK1554: the pair fixture's commits carry a subject and no body, so a flag about a commit body cannot be read there
- 2026-09-06 · RK1553: no sweep asks of a served value flag what RK1517 asks of a boolean, and there are a hundred and twenty-two
- 2026-09-06 · RK1552: the leads a deleted design quoted are read once per path, though the record the departure unpacks carries them
- 2026-09-06 · RK1551: the gate falls silent where a design quotes a constraint and the run that fell silent never says so
- 2026-09-06 · RK1550: a design keeps naming a function of this package after a rename, no gate reading that pointer
- 2026-09-06 · RK1549: no verb removes the vendored engine, so the report that names it hands the last step back to the reader in English
- 2026-09-06 · RK1548: a composed door whose placeholder holds a space is printed unquoted, so the command as printed cannot be run
- 2026-09-06 · RK1547: a pause carries a reason with no date and no expiry, so nothing ever asks whether it still holds
- 2026-09-06 · RK1546: a folded criterion cites no origin, so a reader of the absorbing task cannot tell it was somebody's line
- 2026-09-06 · RK1545: the filings axis reads one per ship over one commit here, which is this session's cadence and not the work's
- 2026-09-06 · RK1544: the brief an estimate prices has no deps and no design, and the row does not say so
- 2026-09-06 · RK1543: a version here names one commit and the surfaces that consume one speak of it as a release, with nothing saying which
- 2026-09-06 · RK1542: the retirement prefix every ranking figure is measured through is split by hand in two tests
- 2026-09-06 · RK1541: the served budget has no room for a tenth subject, and which of its sixteen arguments callers use is unmeasured
- 2026-09-06 · RK1540: a reference page now opens with two lines addressed to the gate, and nothing prices them
- 2026-09-06 · RK1539: the comment explaining the respelling guard names a verb this CLI does not have
- 2026-09-06 · RK1538: the flag saying which ceiling refused a field reaches neither the remedy table nor a payload
- 2026-09-06 · RK1537: a pause is not held to the why limit a project declared, and nothing at the number says so
- 2026-09-06 · RK1536: a decision can only be filed by a departure, so the moment its answer is lost is the moment nothing can be done
- 2026-09-06 · RK1535: nothing records which volunteered rows an author acted on, so the population that could score the read has none
- 2026-09-06 · RK1534: the orientation an install prints names five commands in an order the tree it is printed to cannot run
- 2026-09-06 · RK1533: govern writes a number the config parser then refuses, leaving every verb unable to read the file
- 2026-09-06 · RK1532: thirty rows of the composer work-list share one reason, so none of them says what it would cost
- 2026-09-06 · RK1531: a section title takes the mangled bytes the line's own fields refuse
- 2026-09-06 · RK1530: the corpus reading a refusal was drawn from lives in a docstring and nothing re-takes it
- 2026-09-06 · RK1529: an incidental path naming no file is silent, where a budget naming none is a finding
- 2026-09-06 · RK1528: the near row offers the door to the delivered half and none to the open half it now counts
- 2026-09-06 · RK1527: the near window stayed at three when the corpus it ranks over doubled
- 2026-09-06 · RK1526: the read that lets a reader choose between four rows is repeated on each of them
- 2026-09-06 · RK1525: the kinds sweep is total over the notes that make a call and silent over the ones that do not
- 2026-09-06 · RK1524: the four notes this server appends to a tool result are priced by nothing, on a heavier cadence than the gate's
- 2026-09-06 · RK1523: a declaration whose program engines cannot name is silent, and reads exactly like a project that declares none
- 2026-09-06 · RK1522: one record is a file section, a reference page and a note row, and its docstring describes only the first
- 2026-09-06 · RK1521: the note cadence prices what fired and cannot say what it left out, no list of note codes existing
- 2026-09-06 · RK1520: the marker-leak gate rewards the repair that renders the same wrong bytes
- 2026-09-06 · RK1519: a project that declares its own open markers cannot claim a line, and nothing anywhere says so
- 2026-09-06 · RK1518: adopt refuses two answers inside its estimator, so the dispatcher and the served surface read them as compatible
- 2026-09-06 · RK1517: a served flag that only shapes the terminal form is inert, and nothing asks that of the eighteen
- 2026-09-06 · RK1498: thirty of the thirty-six sites that compose a door are accounted for as a work-list, so nothing ever runs one
- 2026-09-05 · RK1516: section drop deletes a design and never says the constraint answer went with it
- 2026-09-05 · RK1515: a design quoting a constraint to describe somebody else's case reads as having settled it
- 2026-09-05 · RK1514: uninstall leaves the vendored engine on disk and its own kept field never mentions it
- 2026-09-05 · RK1513: brief hands over a line carrying no criteria in silence, and the absence is reported only once part landed
- 2026-09-05 · RK1512: pick and brief never name the deferred store, so a set-aside line and the reason it cites go unread
- 2026-09-05 · RK1511: a line filed because the task that found it could not fit it has no door back into that task's criteria
- 2026-09-05 · RK1510: no read says how many lines a shipped task filed behind it, so granularity is judged from a feel
- 2026-09-05 · RK1509: an adopting project is never shown what its briefs cost, so a ceiling is chosen with no reading in front of it
- 2026-09-05 · RK1508: which engine wrote a project's surfaces is derivable from their bytes and is only ever read out of a record
- 2026-09-05 · RK1507: a line this tool writes under a ledger entry is composed by one function and recognised by another, matching on a prefix
- 2026-09-05 · RK1506: budget prices one line per departure and a ship writes three, so the other two are exemptions rather than answers
- 2026-09-05 · RK1505: the note about a missing page says it is missing and not what it would have told the reader
- 2026-09-05 · RK1504: a served tool name that is also a CLI verb of a different act is a collision nothing enumerates
- 2026-09-05 · RK1503: a field refused by the line quotes a ceiling no config declares, so the reader shortens the prose that fits
- 2026-09-05 · RK1502: a write that composes a prose field can render it without measuring it, and nothing enumerates which ones do
- 2026-09-05 · RK1501: the clause that settles a constraint is deleted by the ship, so the answer lasts exactly as long as the work
- 2026-09-05 · RK1500: the query side of the duplicate ranking cannot be measured, every known answer being a note written after the answer
- 2026-09-05 · RK1499: whether a governed number has a gate behind it is a flag typed at the reading, and no test holds it against the findings
- 2026-09-05 · RK1497: nothing catches the bytes a wrong decode leaves, so mojibake reaches a permanent file and a correction is what cleans it
- 2026-09-05 · RK1496: the filter that makes unclosed honest is inert here, where a hook touches three ungoverned files every commit
- 2026-09-05 · RK1495: the near rows an add prints come from the ledger alone, so two sessions filing one defect never see each other
- 2026-09-05 · RK1494: one gate code now carries four causes with four remedies, so explain can only describe their union
- 2026-09-05 · RK1493: the served notes are four kinds keyed by string literals invented at each call site, and nothing enumerates them
- 2026-09-05 · RK1492: the invoke line finds the program by a .py suffix, which is a guess about a file this tool merges into
- 2026-09-05 · RK1491: the note a wired project reads on every turn is 475 characters at full length and no cadence prices it
- 2026-09-05 · RK1490: a caller who can see a withheld line is worth starting still has no way to take it, only to disbelieve the refusal
- 2026-09-05 · RK1489: the pair sweep reads a flag as swallowed where its fixture cannot hold the state the flag is about
- 2026-09-05 · RK1488: a ship deletes the clause that answered a constraint and says nothing, so the answer ages out unremarked
- 2026-09-05 · RK1487: a vendor that lands and then fails to wire leaves a copy nothing points at, and no refusal says it is there
- 2026-09-02 · RK1486: the brief ceiling was argued from a backlog with no deps at all, so the lists that grow with the graph are outside it
- 2026-09-02 · RK1485: the record that stops a downgrade arrives on the write it guards, so an already-wired project is unprotected until then
- 2026-09-02 · RK1484: a ship recording a checked criterion wraps the entry, so correcting its sentence costs a span this tool wrote
- 2026-09-02 · RK1483: budget and the writes it prices disagree about arguments in both directions, and nothing enumerates the pairs
- 2026-09-02 · RK1482: a finding about the reader's own stale tooling sits in lint's list at the same weight as a line over its limit
- 2026-09-02 · RK1481: the CLI and the MCP surface spell several verbs and flags differently, so a call copied between them is refused
- 2026-09-02 · RK1480: amend replaces the whole dep group, so adding one means retyping every dep the line already carries
- 2026-09-02 · RK1479: a pause's reason is the one departure budget cannot price, its field holding a wrapper and the design carried forward
- 2026-09-02 · RK1478: a settled constraint-line pair is a sentence in one design and no read names it, so the gate's silence has no witness
- 2026-09-02 · RK1477: a re-filing of shipped work in other words ranks 7th, outside both windows, so the add that wrote it agreed it was new
- 2026-09-02 · RK1476: an unscoped list over a ledger is bounded by nothing, so the transport refuses an answer this tool already composed
- 2026-09-02 · RK1475: the ship that finishes a block offers a block drop the decisions file will refuse, on every project that files decisions
- 2026-09-02 · RK1474: a decision's symptom is a copy of a roadmap line that is gone, and no verb in any of the three files can correct it
- 2026-09-02 · RK1473: `unclosed` counts roadkeep's own prose edits as evidence a line shipped, so its report is mostly its own writes
- 2026-09-02 · RK1471: a write from an engine whose home was swapped is allowed, on rules no disk holds and with no pin to compare against
- 2026-09-02 · RK1470: the served staleness note reads mtimes, so an engine replaced under the session looks like somebody saving a file
- 2026-09-02 · RK1469: engines --invoke restates the launcher's order and knows two of its four entries, so a ROADKEEP_HOME pin is missed
- 2026-09-02 · RK1468: the gate's engine note reads the plugin alone, so a project running two local copies at two versions passes it clean
- 2026-09-02 · RK1467: A requirement gates the whole line, so the half of a task that needs nothing is never offered
- 2026-09-02 · RK1466: a project numbering by id gets every address in one answer, which at this repository's size is 961 rows and 175 KB
- 2026-09-02 · RK1465: the launcher spends a whole extra Python start proving an engine runs, doubling how long connecting takes
- 2026-09-02 · RK1464: install --vendor writes the surfaces before it replaces the engine, so one run leaves the outgoing engine's copy on disk
- 2026-09-02 · RK1463: brief spends four fifths of its answer on deps that are all shipped and settled long ago
- 2026-09-02 · RK1462: install rewrites a surface with an older engine's copy, so the remedy lint names undoes a fix that shipped
- 2026-09-02 · RK1461: budget cannot price a line that will carry a requires, so the number it quotes is 21 characters too generous
- 2026-09-02 · RK1460: A criterion checked by running something ships looking exactly like one nobody looked at
- 2026-09-02 · RK1459: The read that prices a body and the write that files it name the same argument differently, so the loop costs a refusal
- 2026-09-01 · RK1458: budget prices a why against the roadmap line's limit while the ship about to be written is held to the ledger's
- 2026-09-01 · RK1457: non-goal.reaches fires on every lint and nothing clears it, so a project that decided reads the same note forever
- 2026-09-01 · RK1455: a changelog listing cannot be asked for its blocks, so the only way to see them is to read the file
- 2026-09-01 · RK1454: block drop refuses a finished block without naming the ending that is correct
- 2026-09-01 · RK1453: A decision's sentence has no correction door, so a typo outlives the decision
- 2026-09-01 · RK1452: the server reports the version it loaded while that same path holds another, and calls the result agreed
- 2026-09-01 · RK1451: engines lists only the copy it is, so a vendored engine the launcher runs has no row and both copies exit clean
- 2026-08-31 · RK1450: anchors lists no rows while its own payload counts 88 of them, and --claims, which narrows it, lists 29
- 2026-08-31 · RK1449: the MCP server does not finish connecting inside the client's 30s timeout, so a session gets none of its tools
- 2026-08-31 · RK1448: a test unpacking report.notes as a whole passes or fails on which unrelated note the gate happened to add
- 2026-08-31 · RK1447: install prints what the surfaces let a session do and its payload says only which files moved
- 2026-08-31 · RK1446: the launcher's `mcp` mode exits 0 and serves nothing on Windows, so a session loses the tools its own hook names
- 2026-08-31 · RK1445: the gate on reads the skill may name lets a piped span through, so the two verbs nothing serves are named anyway
- 2026-08-31 · RK1444: the area prices every surface a session is handed except the skill, and its page has no word left to say so
- 2026-08-31 · RK1443: Every MCP write repeats the same staleness notice a session cannot act on
- 2026-08-31 · RK1442: stats is the one counting verb no tool serves, so the split it just grew is unreachable from the surface an agent has
- 2026-08-31 · RK1441: the --have help names a (needs: ...) group the parser does not read, so a caller copying it is refused
- 2026-08-31 · RK1440: engine.disagreement compares the gate to the plugin and never says the gate is a working tree rather than a release
- 2026-08-31 · RK1439: pick offers a line that seven shipped entries were filed against, with nothing saying it is being worked by proxy
- 2026-08-31 · RK1438: install reports which surfaces it wrote and never says what they let a session do
- 2026-08-31 · RK1437: the skill names all 44 verbs and costs 65k units a turn, so it is a reference loaded as an orientation
- 2026-08-30 · RK1436: amend lists the fields it can change and never names the verb that owns the one it cannot
- 2026-08-30 · RK1435: a refusal diagnoses the write it rejected and never names the read-only verb that would have predicted it
- 2026-08-30 · RK1434: a non-goal can forbid a line the same file lists as ready, and nothing compares the two
- 2026-08-30 · RK1433: nothing checks that a partially-shipped line has criteria, which is the one state where how much is left gets asked
- 2026-08-30 · RK1432: stats counts every open line together, so a reader cannot see how many can actually be started
- 2026-08-28 · RK1431: A page whose sections a component renders publishes an empty On this page, so nothing links them
- 2026-08-28 · RK1430: The walkthrough's commands run off their block, the page scrolling sideways under them
- 2026-08-28 · RK1429: The one slash command that writes puts its argument into a shell unquoted, and nothing holds how any of them does it
- 2026-08-28 · RK1428: Nothing counts the guard's denial, and 1,760 of its 4,092 units re-spell the shell for a caller that has the tools
- 2026-08-28 · RK1427: The release workflow restates the list of files that carry the version, and nothing holds the two together
- 2026-08-28 · RK1426: A finding the gate files with no line loses its doors in the editor, which offers only the explanation
- 2026-08-28 · RK1425: The editor offers every complete door of a finding as an independent quick fix, including a sequence's second step
- 2026-08-28 · RK1424: The skill is 65,180 code units and nothing counts it, where the surface it is larger than has a ceiling
- 2026-08-28 · RK1423: The cost --session row puts the whole surface beside a room derived from a smaller figure
- 2026-08-28 · RK1422: Three of the nine exits the census calls a verdict have nothing asserting the stderr they close with
- 2026-08-28 · RK1421: Nothing enumerates the thirteen exits that return the gate's code, so which is a verdict is settled one report at a time
- 2026-08-28 · RK1420: install --check closes a complete answer with the capture offer, on the verb an adopter runs while wiring
- 2026-08-28 · RK1419: The capture offer rides lint --fix and repair, whose exit 1 is the verdict lint's own is exempt from
- 2026-08-28 · RK1418: ref.missing asks the author for an anchor on the one scheme that derives it
- 2026-08-28 · RK1417: The pitch is a hand-edited HTML file inside the governed store, which is also the web root the site is served from
- 2026-08-28 · RK1416: The decision field explain publishes is the same string as the cause on every code that carries one
- 2026-08-28 · RK1415: A project whose roadmap already exists cannot reach a first roadkeep.toml through any verb
- 2026-08-28 · RK1414: The order of the reference pages is a number typed into each one, on pages whose every other fact is derived
- 2026-08-28 · RK1413: Most finding codes have no hand-written situation, so their page classifies the error and never names it
- 2026-08-28 · RK1412: The area's build runs on changes to site and docs, and every page it generates is read out of src
- 2026-08-28 · RK1411: lint died with a RuntimeError traceback on a machine whose environment names no home directory
- 2026-08-28 · RK1410: An area published as HTML alone makes an agent render a page to read prose the repository wrote as Markdown
- 2026-08-28 · RK1409: Nothing bounds a page, in the one project that measured what a file with room on every page grows into
- 2026-08-28 · RK1408: Prose four files own gets told again by any page that frames it, and the copy nobody reads is the one that drifts
- 2026-08-28 · RK1407: The model the verbs act on is written for an agent mid-session and for whoever edits the code, never for a reader
- 2026-08-28 · RK1406: Adoption on a repository that already has a backlog is a list of commands with no run and no refusals shown
- 2026-08-28 · RK1405: Nothing describes what a session receives once the plugin is installed, only how to install it
- 2026-08-28 · RK1404: Writing a first configuration file needs the read that describes it, which needs a configured tree to answer
- 2026-08-28 · RK1403: A finding code pasted into a search engine resolves to nothing, and explain answers only from an installed copy
- 2026-08-28 · RK1402: Nothing outside a checkout says what a verb takes, so a reader picks flags by installing the tool first
- 2026-08-28 · RK1400: No job builds the area, so a page that stopped compiling is found by a reader and a stale commit is published
- 2026-08-28 · RK1399: The directory GitHub Pages serves is the governed store, so a build writing into it lands on the roadmap
- 2026-08-28 · RK1398: An adopter's questions are answered by one long README or by a command they have not installed yet
- 2026-08-27 · RK1401: No read prints the parser as data, so anything outside a terminal retypes the verbs and their flags
- 2026-08-27 · RK1397: The module that declares the capture family says two verbs have this tool as their subject, and it declares four
- 2026-08-27 · RK1396: A path argument ignores -C, so a door this tool prints is refused as a file that does not exist
- 2026-08-27 · RK1395: A capture filed upstream reads as unfiled in two reports, and neither names the stamp that would resolve it
- 2026-08-27 · RK1394: nothing deletes a capture whose task shipped, so the reports directory grows for ever and is cleared by hand
- 2026-08-26 · RK1393: the authority on the query surface does not name the subject config gained, so an agent cannot discover it there
- 2026-08-26 · RK1392: engines reads a fourth copy and every sentence that counts them still says three
- 2026-08-26 · RK1391: a frozen fixture nobody declared is read by nothing and reported by nothing, though the table says otherwise
- 2026-08-26 · RK1390: a third shape the re-pin retired has a test and no frozen fixture, so it is read off no backlog at all
- 2026-08-26 · RK1389: merge --check offers one copy where the register report names both, and the other reads now send the caller here
- 2026-08-26 · RK1388: two reads answer about one half of the wiring each and neither says which, so together they read as a contradiction
- 2026-08-26 · RK1387: install --check offers to wire a driver the files already route to, so two reads disagree about one fact
- 2026-08-26 · RK1386: register wires whatever installed the console script, so a developer's checkout cannot be the driver its own merges run
- 2026-08-26 · RK1385: the driver git runs is a fourth copy of this tool and no read says which, so a build behind the tree merges its files
- 2026-08-26 · RK1384: this repository ships a merge driver and routes none of its own files to it, so a merge here falls back to git's markers
- 2026-08-26 · RK1383: twenty-one pairs of modules import each other and nothing names them, so a new cycle looks like the ones already there
- 2026-08-26 · RK1382: the record a reading returns is declared in the module that prints it, so the two import each other at call time
- 2026-08-26 · RK1381: the conversion fixed from this package's own corpus has no read that states the reading, so its drift needs a script
- 2026-08-26 · RK1380: a law a governed sentence cites is resolved by nothing, so a number outside the six reads as a law
- 2026-08-26 · RK1379: five of the eight blocks closed with nothing stating what finished them, which is the emptiness RK1358 says out loud
- 2026-08-26 · RK1378: the refusal that will not move an anchor across parents names the family and not the free address under it
- 2026-08-26 · RK1377: four of the six writes the sweep excludes are excluded for the fixture's shape and not for anything about the verb
- 2026-08-26 · RK1376: the one-column claim a ledger entry makes about every write is held by a test that drives one verb
- 2026-08-26 · RK1375: brief states one field's allowance in two grammars on adjacent rows, so the comparison they exist for is read twice
- 2026-08-26 · RK1374: the volunteered rows are bounded at three and say neither how many of how many nor where the rest are
- 2026-08-26 · RK1373: the neighbours read guards on a path existing where its siblings guard on a file, so a role naming a folder fails an add
- 2026-08-26 · RK1372: the add answer prints its own rows at two columns, so one verb's report is two grammars a reader has to hold
- 2026-08-26 · RK1371: every reason for not serving an argument is declared and read by no command, so a caller asking why reaches none
- 2026-08-26 · RK1370: an add answers without the delivered entries nearest what it filed, so a duplicate is found after the design is written
- 2026-08-26 · RK1369: the check that unoffered arguments say why skips any tool an always flag made, so claim withheld --have and nothing said
- 2026-08-26 · RK1368: the staleness note rides on a refusal, so a write validated against code that changed on disk is answered with nothing
- 2026-08-26 · RK1367: govern only stacks, so a number whose premise was falsified keeps every paragraph arguing for the reading it replaced
- 2026-08-26 · RK1366: budget states the room left beside a why no write extends, so an amend is told 8 words where it may write 31
- 2026-08-26 · RK1365: brief reports 37 characters left for a ship why and then accepts one of 145
- 2026-08-26 · RK1364: the session ceiling is declared against a package-wide surface no project is sent, so it bounds nothing in particular
- 2026-08-26 · RK1363: an outline project cannot give a decision a body, the id being no address in a file that numbers its own headings
- 2026-08-26 · RK1362: the spec an agent executes is assembled per call by brief and exists nowhere a review or a second tool can read
- 2026-08-26 · RK1361: a decision is one sentence, so the alternatives rejected and the consequences accepted are kept by hand or not at all
- 2026-08-26 · RK1360: a project declaring no decisions file still pays for its verbs at connect, the tool list being built from the parser
- 2026-08-25 · RK1359: the install report heads with the engine checkout and never names the project it wrote to
- 2026-08-24 · RK1358: a block that finishes with no criteria reads like one that met them, where the project declared criteria
- 2026-08-24 · RK1357: the queue's finding claims a task line its own verbs decline, so the gate and the read disagree
- 2026-08-24 · RK1356: the criteria reader claims a task line as its own, and the drop it prints deletes the line
- 2026-08-24 · RK1355: a task line under Non-goals is read as both, and the door the non-goal finding prints deletes it
- 2026-08-24 · RK1354: the driver sees a composed dangling dep and not a composed cycle, which the same backlog decides
- 2026-08-24 · RK1353: a merge composes a dep on a line the other side removed and returns it clean, the gate it runs being per-file
- 2026-08-24 · RK1352: the merge driver refuses over a defect its inputs already carried, so a branch is blocked by a line it never wrote
- 2026-08-24 · RK1351: the estimate lists no byte-level code, so a BOM the fixer closes is missing from what adoption costs
- 2026-08-24 · RK1350: adopt on a file that is not text prints the decoder's own message, naming no file, no verb and no door
- 2026-08-24 · RK1349: adopt reports a field the ledger declaration removed, so a row that cannot exist prints 0 over
- 2026-08-24 · RK1348: adopt counts prose lines over a width nothing refuses, so the estimate prices work no gate would ask for
- 2026-08-24 · RK1347: the reread covers two roles of three, so a prose file is priced at nothing and labelled a backlog
- 2026-08-24 · RK1346: adopt calls 361 changelog lines changing where --ledger reads all of them as conforming, and names the flag nowhere
- 2026-08-24 · RK1345: adopt prints every field as longest 0 and 0 over where nothing parsed, which reads as fields that fit
- 2026-08-24 · RK1344: two adopt gains have labels the fixed pad cannot separate, so the row runs its first word into its sentence
- 2026-08-24 · RK1343: govern tells every undeclared key that a build default applies, and three of them have no default at all
- 2026-08-24 · RK1342: deps prints its own spelling of no-open-task, so a paused line is invisible where the shared refusal names it
- 2026-08-24 · RK1341: a paused id is refused as never written or retired, when the store it sits in has a read that prints it
- 2026-08-24 · RK1340: a door prints a blank for the id its own finding carries, so the reader types what the tool just held
- 2026-08-24 · RK1339: filled drops a positional blank instead of failing loudly, so seven doors would run as a different command
- 2026-08-24 · RK1338: the door-execution test runs one remedy row of 118, so a door naming a refused command is found by hand
- 2026-08-24 · RK1337: the duplicate-anchor remedy names a command that exits 2 in the very state that produces the finding
- 2026-08-24 · RK1336: the session budget's door reprints the total instead of ranking the descriptions a reader has to cut
- 2026-08-24 · RK1335: cost --tools states a total the gate no longer uses and no session ceiling, on the read used to decide a cut
- 2026-08-24 · RK1334: the served surface is measured with the handshake's provenance in it, so the budget moves with the checkout
- 2026-08-24 · RK1333: cost --session prints the tools total without the ceiling it is measured against, in both registers
- 2026-08-24 · RK1332: every design clause respells an anchor the entry already carries, in the one sentence proven to have no room
- 2026-08-24 · RK1331: the room a crowded recording names is the outcome plus the note, but it is reported as the outcome alone
- 2026-08-24 · RK1330: the supersession trace is the clause that gives way, so a ship recording where a design went cannot say it was stale
- 2026-08-24 · RK1329: a section written with its id as the title reads §RK1320 RK1320, and nine of nine filings did it
- 2026-08-24 · RK1328: a project past init opens an opt-in table only by hand-editing the config this tool owns the writes to
- 2026-08-24 · RK1327: a repair --fix makes with no word in REPAIRS goes unnamed in all six copies, and the gate stays green
- 2026-08-24 · RK1325: a project wiring no plugin is told its doors arrive under one, so a payload names a tool nothing there answers
- 2026-08-24 · RK1324: four key names carry the same runnable command across the payloads, so a consumer needs four readers for one idea
- 2026-08-24 · RK1323: this repository declares no [criteria], so the fixture that proves the format does not exercise a table init now writes
- 2026-08-24 · RK1322: the tree is split between two line terminators and nothing reads which a file is, so an append lands in the wrong one
- 2026-08-24 · RK1321: budget serves seven subjects on one tool and is now the largest, so every subject added re-argues a shared ceiling
- 2026-08-24 · RK1320: one drafted flag answers two ways about the same prose, so a number is labelled the caller's on a line the file holds
- 2026-08-24 · RK1318: the gate reads a criterion's shape and length and never whether the block or id it is addressed to still exists
- 2026-08-23 · RK1319: the event says a block finished and not that its heading may go, though the offer now depends on config it omits
- 2026-08-23 · RK1317: `renumber` moves a line, its section and its deps, and leaves its `## Done when` list on the number it spent
- 2026-08-23 · RK1316: `block drop` withdraws a label and leaves its `## Done when` list addressed to a block no file declares
- 2026-08-23 · RK1315: weigh passes every ledger commit as one argument, so git refuses the call once the ledger is long enough
- 2026-08-23 · RK1314: config prints the criteria table under the non-goals sentence, so the read names the wrong table
- 2026-08-23 · RK1313: init writes no [criteria] and no [requirements], so the verbs that fill them refuse on a fresh project
- 2026-08-23 · RK1312: section amend --replace refuses a sentence the section carries, because the stored copy is wrapped across lines
- 2026-08-23 · RK1311: amend --requires alone is refused as nothing to amend, then reports unchanged on the call that writes it
- 2026-08-23 · RK1310: nothing resolves a sentence to the anchor carrying it, so a section amend addressed by text is guessed and retried
- 2026-08-23 · RK1309: a new line's section budget is reachable only through brief, which needs an id, so the first body is written blind
- 2026-08-23 · RK1308: lint exits 1 on install.stale, so a CI gate goes red on a backlog it has just reported clean
- 2026-08-23 · RK1307: The json payload drops the sentence and the remedy the human output carries, so an MCP caller gets less than the CLI
- 2026-08-23 · RK1306: brief states the ledger allowance but not what --recorded-in and --superseded-design spend from it
- 2026-08-23 · RK1305: budget answers for a line and a ship and not for a retirement, so its reason is drafted blind
- 2026-08-23 · RK1304: brief says the priority queue names nothing ready and picks elsewhere, without naming the task that would make it ready
- 2026-08-23 · RK1302: ship --part leaves a line in the in-progress tier, so pick keeps offering a remainder that is waiting on unshipped work
- 2026-08-23 · RK1301: brief lists every id the task transitively unblocks, so the earliest line in a graph answers with 79 ids nobody reads
- 2026-08-23 · RK1300: the ship that finishes a block says finished and does not carry the criteria that decide whether it is
- 2026-08-23 · RK1299: a clean lint answers with 42 note rows repeating six facts, five times the cost of its text form
- 2026-08-23 · RK1298: brief answers with two budget objects that differ in six fields and repeat every other row
- 2026-08-21 · RK1297: a line no caller in this checkout can execute reads as ready, so `pick` offers it again every call
- 2026-08-21 · RK1296: the argument a number carries is kept in the file and never returned by the read that asks about the number
- 2026-08-21 · RK1295: an argument that is only whitespace is reported as written above the key and is not
- 2026-08-21 · RK1294: a re-run stacks the argument it already placed, so the same sentence appears twice above one number
- 2026-08-21 · RK1293: the argument for a limit is ruled into the commit body, which a tool this project does not own composes
- 2026-08-20 · RK1292: the header says nothing is over the ceiling while the listing under it names a line nobody could measure
- 2026-08-20 · RK1291: the read prices open lines and prices whatever it is named, so a shipped id answers a figure comparable to nothing
- 2026-08-20 · RK1290: the note saying what the gate did not price fires on every clean run of any real backlog
- 2026-08-20 · RK1289: the note's total is the backlog minus whatever would not compose, so it understates by the lines it warns about
- 2026-08-20 · RK1288: a line whose brief will not compose is dropped from the ranking in silence, and the widest may be it
- 2026-08-20 · RK1287: the gate composes one brief per open line, so a project that declared a ceiling pays it on every commit
- 2026-08-20 · RK1286: the read that exists to fit in a tool result is the one thing here with no budget, and it grew four rows
- 2026-08-20 · RK1285: a refusal prints a door from the field beside one from the case, and the two say opposite things
- 2026-08-20 · RK1284: the reading behind a decisions limit sees only that file, so a number it accepts refuses every later ship
- 2026-08-20 · RK1283: the guard parses the config twice on every allowed write, once to refuse nothing and once to advise
- 2026-08-20 · RK1282: the shape prints one declared value for a table the project declared several times, and says nothing about the rest
- 2026-08-20 · RK1281: a decision inherits the roadmap claim, so a tighter limit on that file refuses a ship over a field no flag writes
- 2026-08-20 · RK1280: the guard allows a hand edit to the config and says nothing, so a limit with a door is still typed past it
- 2026-08-20 · RK1279: the reading behind a limit walks a hand-written list of roles, so a role added later is measured by nobody
- 2026-08-20 · RK1278: the config shape prints the default beside a key the project set, so the number shown is not the one in use
- 2026-08-20 · RK1277: the config completion caches a fact about the file on the engine's clock, so a declared key still reads as undeclared
- 2026-08-20 · RK1276: a criteria address refused for a paused id says the work shipped, so a pause reads like a typo
- 2026-08-20 · RK1275: the shipping allowance names one of the three writes a ship composes, so two sentences are written blind
- 2026-08-20 · RK1274: a decision is never marked superseded, so the file records that one was made and not that it stopped holding
- 2026-08-20 · RK1273: a served field's description states one rule twice, so the ceiling is spent on a paragraph nobody reads twice
- 2026-08-20 · RK1272: `[limits]`, `[budgets]` and `[markers]` have no verb, so the file governing every write is the ungoverned one
- 2026-08-20 · RK1271: the editor completes nothing in `roadkeep.toml`, so every key is typed from memory
- 2026-08-20 · RK1270: nothing prints what `roadkeep.toml` may declare, so its keys live only in the parser that rejects them
- 2026-08-20 · RK1269: a decision that outlives the work explaining it has no governed file, so an ADR is kept by hand or not at all
- 2026-08-20 · RK1268: a definition of done is addressed by block, so what finishes one task is written nowhere a verb reads
- 2026-08-20 · RK1267: ship deletes the design and nothing records where the reasoning that outlives the code went
- 2026-08-20 · RK1266: four non-goal codes report a defect and name no command that closes it
- 2026-08-20 · RK1265: a non-goal states what is not built, and nothing states what must be true for a block to be finished
- 2026-08-20 · RK1264: the scaffold is the only writer of [files], so a project past init cannot declare a role it turns out to want
- 2026-08-19 · RK1263: section amend replaces the whole body, so fixing one citation means re-emitting every table and code fence verbatim
- 2026-08-19 · RK1262: a refused --remainder is reported as a why, so the retry has to guess which of the two strings was short a full stop
- 2026-08-19 · RK1261: a ship refusal blames --why for characters that came from --superseded-design
- 2026-08-19 · RK1260: the MCP schemas offer '-' and stdin, an input channel no MCP client can provide
- 2026-08-19 · RK1259: adoption writes no deferred store, so the first defer refuses at the moment its reason is already written
- 2026-08-19 · RK1258: a block's first task files its rationale AS the family heading; every later one gets a child, and no verb picks
- 2026-08-18 · RK1256: A structural refusal and a schema refusal never appear together, so a call wrong twice is refused twice
- 2026-08-18 · RK1255: A refusal names one violation while the next is already true, so one sentence costs several round trips
- 2026-08-18 · RK1254: A positional spelled as a flag is answered with a list of flags that cannot contain it
- 2026-08-18 · RK1253: The total whose whole purpose is comparison is the one with no section breakdown
- 2026-08-17 · RK1252: The breakdown that says where to cut ranks by a unit the total above it is no longer in
- 2026-08-17 · RK1251: A declared file that is not on disk is reported as one this tool could not decode
- 2026-08-17 · RK1250: The read consulted before an edit lacks the figure the read consulted after it now has
- 2026-08-17 · RK1249: The summary row claims bytes is what the gate refuses on while the row above it names lines
- 2026-08-17 · RK1248: The one row whose ceiling this package hardcodes states its room and the rows the project declared do not
- 2026-08-17 · RK1247: The one message that stops an Edit defaults its prefix to a route the session may not have
- 2026-08-17 · RK1246: Four messages carry the served prefix from four sites and nothing holds the four together
- 2026-08-17 · RK1245: The read that exists to compare two costs prints them in two different units
- 2026-08-17 · RK1244: A three-way answer is carried as two fields, so a Notice can be built stating a pair that cannot happen
- 2026-08-17 · RK1243: The line every session is handed has a budget nothing reads and a constant somebody edits
- 2026-08-17 · RK1242: The MCP tools the session-start note names never arrive, so the agent falls back to the CLI
- 2026-08-17 · RK1241: The description split does not name its own unaccounted bytes where the tool total does
- 2026-08-17 · RK1240: One pinned key silences a finding and enables two refusals, added by three tasks reading it three ways
- 2026-08-17 · RK1239: The description row is the largest a tool has and nothing says which of its parts to shorten
- 2026-08-17 · RK1238: The pen is guarded against a stale copy and the gate that judges it is the same stale copy
- 2026-08-17 · RK1237: The pin guard runs three git calls in front of every write and nobody measured them
- 2026-08-17 · RK1236: The tool over its ceiling is named and which of its arguments spent the bytes is not
- 2026-08-17 · RK1235: A copy behind the wired one still writes, and the note that says so is read after the line has landed
- 2026-08-17 · RK1234: task.worked fires per line, so the block that quietly finished is still only visible one commit at a time
- 2026-08-17 · RK1233: The open half of a partial is still an inference the reader makes, where the landed half is recorded
- 2026-08-17 · RK1232: A full run asks for every core, so the machine has nothing left while the suite is running
- 2026-08-17 · RK1231: renumber leaves the section heading it wrote, so an outline-ref project keeps a title naming the old id
- 2026-08-17 · RK1230: Nothing tells a shell caller which engine copy is the one wired to this project
- 2026-08-17 · RK1229: A dep with a parenthesis is written, then no verb can reach the line it made unreadable
- 2026-08-17 · RK1228: Nothing reports source changed under an open task while the line stayed open
- 2026-08-17 · RK1227: section amend accepts prose citing an anchor that does not exist, so a docs-only commit turns an adopter's gate red
- 2026-08-17 · RK1226: Nothing names the open half of a partial, only the half that shipped
- 2026-08-17 · RK1225: budget publishes every width a field has and not the one sentence it accepts
- 2026-08-17 · RK1224: A refusal on one over-long field discards the section body sent with it, so the retry re-sends every word
- 2026-08-17 · RK1223: block add says nothing to open when the label is in the ledger but the roadmap, where the work goes, has none
- 2026-08-17 · RK1222: The gate reads a declared query for a grammar it cannot parse and never for a pathspec that answers about nothing
- 2026-08-17 · RK1221: budget on an id the roadmap holds reads the four fields describing a line and discards them without a word
- 2026-08-17 · RK1220: A refusal preamble is spelled with the invocation, so the test that walks the stair it prints reads it as a step
- 2026-08-17 · RK1218: add cannot carry the section it requires, so every filing is two commands with a dangling pointer between them
- 2026-08-17 · RK1217: path.missing judges a shipped entry against today's tree, so a file later moved to another repo makes history a finding
- 2026-08-17 · RK1214: a resolved engine that fails to import takes the whole command down instead of falling through to the next candidate
- 2026-08-17 · RK1209: No test runs the commands this tool composes, so a refusal naming a call that refuses stays green
- 2026-08-17 · RK1206: Under an outline, the ref.unresolved remedy names the task id where the missing section is the anchor the line points at
- 2026-08-17 · RK1203: lint offers amend as the door for a path.missing on a changelog line, and amend refuses every shipped id
- 2026-08-17 · RK1202: guard exits 0 and prints nothing on a payload it could not parse, which is the answer it gives for allowed
- 2026-08-17 · RK1200: The committed launcher reads ROADKEEP_HOME verbatim and lists no candidate inside the repository
- 2026-08-17 · RK1193: Every adopting repository has to write its own installer to pin the engine it runs
- 2026-08-16 · RK1219: Share declares payload twice, so the definition a reader edits may be the one Python has already discarded
- 2026-08-16 · RK1216: remaining reports 0 sites for a pathspec that matched no files, which is what it reports for work that is finished
- 2026-08-16 · RK1215: defer refuses a partially shipped line in the ledger's words about a second entry it would not write
- 2026-08-16 · RK1213: A paused id is refused in the words an id nobody wrote gets, and Whereabouts says no file mentions it
- 2026-08-16 · RK1212: The id-reuse refusal tells a caller to omit --id, and renumber declares no --id
- 2026-08-16 · RK1211: anchors --next on an empty outline names a roman first address by hand, on a project that may number decimally
- 2026-08-16 · RK1210: next_family answers a different free address per process where two numbering systems tie on the same file
- 2026-08-16 · RK1208: section add refuses a child whose family is missing and writes a grandchild whose middle generation is
- 2026-08-16 · RK1207: section add refuses a child whose family does not exist and names no verb, listing what the file declares instead
- 2026-08-16 · RK1205: The section add an add prints to close the pointer it wrote refuses where that anchor's family does not exist
- 2026-08-16 · RK1204: A block title is write-once: nothing renames one, and block drop is refused the moment a line is filed under it
- 2026-08-16 · RK1199: The shipping allowance brief prints is ten characters narrower than the limit ship then enforces
- 2026-08-16 · RK1198: Filing into a block whose prose family is spent is four sequential refusals, each teaching one step
- 2026-08-16 · RK1196: restate replaces a false premise and leaves the why and the design still asserting it, naming no follow-up
- 2026-08-16 · RK1192: A launcher and skill older than the version answering are reported by nothing the gate runs
- 2026-08-16 · RK1190: budget states the allowance and cannot be handed a draft, so prose three words over is found by being refused
- 2026-08-16 · RK1188: No verb says what the blocks are called, so choosing where a task goes means reading the file the hook denies
- 2026-08-16 · RK1187: `symptom` is the one prose field no verb declares as a pipe reader, so `-` lands in the file as the claim
- 2026-08-16 · RK1186: `init` scaffolds three files and no flag names the strategy one, so a spec above the task line has no home
- 2026-08-14 · RK1201: An open line whose work is already committed is invisible until somebody runs origin on that one id
- 2026-08-14 · RK1197: A loose path that is already staged reads the same as one merely edited, so a stale index entry commits silently
- 2026-08-14 · RK1195: Nothing in the suite sees a top-level definition left without its blank-line separator
- 2026-08-14 · RK1194: A local variable with an import's name makes the unspelled-import test read a dead import as used
- 2026-08-14 · RK1191: _add still computes a `piped` flag that nothing reads, left behind when the pipe clash moved to dispatch
- 2026-08-14 · RK1189: A wired plugin stood the committed launcher's MCP server down, so the harness marked that server failed
- 2026-08-14 · RK1185: `brief` hands over a task's design and deps but never its criterion, so it is read at the ship
- 2026-08-14 · RK1184: Nothing states what would prove a task done, so `ship` is a judgement with no artefact behind it
- 2026-08-14 · RK1183: delivered --near cannot reach a partner whose symptom is stated differently, and its closure now carries an exemption
- 2026-08-14 · RK1181: ref.dangling reads a section mark citing another document as a pointer into this file, and the first has no spelling
- 2026-08-14 · RK1180: A block that is a standing category empties whenever it is caught up, and ship reads that as completion
- 2026-08-14 · RK1179: A checkout whose source does not import answers with a raw Python traceback instead of a sentence
- 2026-08-14 · RK1178: anchors recommends an address that was declared and retired inside one commit
- 2026-08-14 · RK1177: add --ref accepts a retired anchor that section add then refuses
- 2026-08-14 · RK1176: ship writes a literal dash where --superseded-design was handed the pipe every other prose argument reads
- 2026-08-14 · RK1172: The gate's checks are functions gathered by convention, each with its own signature, so adding one is invisible
- 2026-08-14 · RK1171: build_parser is one 2,000-line function, so a verb's flags are edited in the file the whole surface is in
- 2026-08-14 · RK1170: A verb's plain answer and its json payload are built in two files, and most of the printing never moved
- 2026-08-14 · RK1169: Six tables in serving.py restate per verb what its own parser already declares
- 2026-08-13 · RK1175: a partial ship's entry is listed as accounted for by no commit, for as long as it stays partial
- 2026-08-13 · RK1174: The why budget brief prints is the roadmap line's, and the ship that follows enforces a different one nothing shows
- 2026-08-13 · RK1168: Declaring a namespace re-addresses the headings and leaves the file's own citations pointing at the other prose file
- 2026-08-13 · RK1167: Two registry rows for one project make the first win, so engines names a version that was replaced and is not on disk
- 2026-08-13 · RK1166: A registry row whose install directory is gone stands the launcher down, and nothing guards the governed files
- 2026-08-13 · RK1165: gaps prints a numbering jump as 499 rows, and the two ids worth reading are behind them
- 2026-08-13 · RK1164: ship does not say what is left in the block, so every iteration of a block run costs a second call
- 2026-08-13 · RK1163: A rationale written before its dep shipped reads as current, and brief is where that is decided
- 2026-08-13 · RK1162: a delivery elsewhere and a local filing are one number in the payload
- 2026-08-13 · RK1161: the delivery stamp asks for the repository the config already declares
- 2026-08-13 · RK1160: A capture filed in another repo's backlog can never clear its row, and only a wrong id or a deletion silences it
- 2026-08-13 · RK1159: nothing here runs the suite the way CI does, so four defects this session were only findable there
- 2026-08-13 · RK1158: an API newer than the supported floor is green on this machine and red only in CI
- 2026-08-13 · RK1157: the declaration and the answer differ by one letter, and the declaration is the reachable one
- 2026-08-13 · RK1155: a test simulates a missing engine by pointing one variable at nothing, and the cache answers instead
- 2026-08-13 · RK1154: an installed engine's name is a substring of the tool name, so three guard tests fail in CI
- 2026-08-13 · RK1153: three tests spawn git around the suite's own runner, so CI has no committer identity
- 2026-08-13 · RK1152: ship refuses because a dependent task's why is one character too long once its dep annotation grows a tick
- 2026-08-13 · RK1151: a citation inside an indented code block is read as a citation, so the gate is red on quoted output
- 2026-08-13 · RK1150: An older installed roadkeep reports a newer config's key as a typo, so the fix reads as deleting it
- 2026-08-13 · RK1149: A refusal that already computed the free anchor still makes the caller retype it
- 2026-08-13 · RK1148: three skips claim a shape left the corpus and it was never there at the pin before
- 2026-08-13 · RK1147: the adopt payload publishes the count of an unread reading and not the flag that changes it
- 2026-08-13 · RK1146: a backlog written before the tool has no live evidence left, so two tests about disagreement only skip
- 2026-08-13 · RK1145: five tests skip because the shape they read left the corpora, so re-pinning retires coverage without saying so
- 2026-08-13 · RK1144: the corpora pins are 477 deliveries behind, so the shapes those projects adopted since are untested here
- 2026-08-13 · RK1143: stats keeps printing the capture row after every capture is filed, so a permanent line says nothing is owed
- 2026-08-13 · RK1142: a capture already on disk has no command that stamps it, so the only door is the Python API
- 2026-08-13 · RK1141: a capture whose defect shipped under a reworded symptom stays unfiled for ever, and nothing can clear the row
- 2026-08-13 · RK1140: anchors --next names a free top-level that add --ref then refuses, because no heading declares it yet
- 2026-08-13 · RK1139: A kept capture is write-only: nothing lists it, counts it or says it is not a backlog line
- 2026-08-13 · RK1138: A partial that is later abandoned has no exit: retire refuses while its entry exists
- 2026-08-13 · RK1137: git collapses an untracked tree to its top directory, so a claim naming a file inside it is called a typo
- 2026-08-13 · RK1136: 26 lines of the every-turn file are needed only on a turn that builds or commits, and every other turn pays for them
- 2026-08-13 · RK1135: agents.md has one line of budget left, so the next surface the index must name has nowhere to be named
- 2026-08-13 · RK1134: the index closure reads the git index, so a new surface reaches it one commit after it was written
- 2026-08-13 · RK1133: amend --json carries which fields changed and never what they were, where status and restate both answer
- 2026-08-13 · RK1132: 45 modules end CRLF and 11 end LF, so an edit anchored on one terminator silently matches nothing
- 2026-08-13 · RK1131: six records gained a field and twelve payloads carry it, with a closure over one dataclass and none over the rest
- 2026-08-13 · RK1130: four of the thirty-two write commands print a staging line, so the rest can leave the projection they refreshed behind
- 2026-08-13 · RK1129: `add` refreshes the README projection and prints no staging line, so a commit takes the roadmap and leaves it behind
- 2026-08-13 · RK1128: A --lines refusal says the ledger records no partial, on a line whose partial it names one call later
- 2026-08-13 · RK1127: a task's own subsection reads as another design, the exclusion comparing labels as strings and not as addresses
- 2026-08-13 · RK1126: a body-only section amend leaves this task's own rationale file unaccounted, so a departure calls it nobody's
- 2026-08-13 · RK1125: a prose file this task wrote hides another session new section, the reading being lines and a rationale holding none
- 2026-08-13 · RK1124: sharing resolves the same revision once per carrier, so a read pays a subprocess for an answer it already has
- 2026-08-13 · RK1123: nothing binds the fields of Scope to the two payloads that carry them, so a field added reaches one reader
- 2026-08-13 · RK1122: `claim <id>` answers an empty shared list on every call, so the read a commit is composed from sees half its contract
- 2026-08-13 · RK1121: the block-empty hint names a door six ships out of nine, in a project that has never dropped a heading
- 2026-08-13 · RK1120: a roadmap carrying two sessions lines is one accounted file, so a departure cannot separate them
- 2026-08-13 · RK1119: the entry-point sentence `install` writes says it wired a checkout, which a `--committed` project has none of
- 2026-08-13 · RK1118: `section show --own` on a container prints an empty body, which reads exactly like a command that failed
- 2026-08-13 · RK1117: `ship` prints a stage line without naming the paths in it that carry changes no claim owns
- 2026-08-13 · RK1116: the launcher the installed skill names as the entry point forwards only guard and mcp, so every verb exits 2
- 2026-08-13 · RK1115: defer refuses any line whose why already fills its budget, because the wrapped reason is charged to the same limit
- 2026-08-13 · RK1114: `pick` offers a partially shipped line that `--claim` then refuses, the partial ledger entry reading as a ship
- 2026-08-13 · RK1113: `install --check` reports drift on a `--committed` project and names the plain `install`, which downgrades the wiring
- 2026-08-13 · RK1112: section show prints the subtree and section amend takes only the own body, so the obvious round-trip is refused
- 2026-08-13 · RK1111: a namespaced citation is read as its namespace, so lint reports a live anchor as dangling
- 2026-08-12 · RK1110: a table of contents lists the file's own headings and nothing re-derives it, so a ship leaves it wrong
- 2026-08-12 · RK1109: section amend answers unchanged: it already reads that way over a body left unread on stdin
- 2026-08-12 · RK1108: the guard reaches no environment without the plugin, so a project hand-writes the launcher that finds one
- 2026-08-12 · RK1107: a prose file's preamble and table of contents carry no anchor, so every edit to them is a hand edit
- 2026-08-12 · RK1106: a citation inside prose can name a section that no longer exists, and no gate code says so
- 2026-08-12 · RK1105: a byte budget counts what the checkout has, so CRLF and LF give one commit different headroom
- 2026-08-11 · RK1104: the rule about guessing a fact from a file's text is held over one file and stated for the rest
- 2026-08-11 · RK1103: adopt resolves a project path from the root and no other verb taking one was checked
- 2026-08-11 · RK1102: a predicate that reads a governed file's text instead of asking the parser has now been wrong twice
- 2026-08-11 · RK1101: a relative path under -C resolves against the shell's directory, so adopt can measure another project's file
- 2026-08-11 · RK1100: adopt estimates what a project gains and never says the tool surface costs it fifty thousand characters
- 2026-08-11 · RK1099: a flag added to a served verb is CLI-only until somebody lists it, and nothing says which ones are
- 2026-08-11 · RK1098: two tests read this repo's docs as a fixture and assume the roadmap has an open line
- 2026-08-11 · RK1097: the served surface is fifty thousand characters against eight thousand the gate holds
- 2026-08-11 · RK1096: budget has six subjects and the two that read a session's cost overlap without sharing a measurement
- 2026-08-11 · RK1095: two reads rank what a session pays for and neither can be asked about the other
- 2026-08-11 · RK1094: the standing advice on which half of agents.md to compress rests on a figure that is off by half
- 2026-08-11 · RK1093: the estimate's four gains are four hand-written blocks in one function
- 2026-08-11 · RK1092: agents.md is eight bytes under its budget and the next sentence has nowhere to come from
- 2026-08-11 · RK1091: three heredoc edits corrupted a test file's string literals and only the parser noticed
- 2026-08-11 · RK1090: the queue is the fourth door an estimate could name and the only member left out
- 2026-08-11 · RK1089: the estimate answers what one file costs and says nothing about the doors a project would gain
- 2026-08-11 · RK1088: the closure names the file it removed from and no caller reads it
- 2026-08-11 · RK1087: the deferred store is a declared feature no adopting corpus has ever turned on
- 2026-08-11 · RK1086: Resumption models a placement so a resume that places nothing has to fake one
- 2026-08-11 · RK1085: the pairwise loop reads a role's whole document per pair and lint already holds both
- 2026-08-11 · RK1084: the pair the declaration says nobody has walked into is reachable and still unread
- 2026-08-11 · RK1083: resume answers two different acts under one verb and the printer cannot tell them apart
- 2026-08-11 · RK1082: the pairwise checks over three governed files are two rules written by hand
- 2026-08-11 · RK1081: an id the roadmap and the deferred store both carry passes the gate and ship leaves it shipped and paused
- 2026-08-11 · RK1080: the gate and the verb decide a live partial in two places that only agree by hand
- 2026-08-11 · RK1079: the door table crosses two axes and the deferred store is a third nobody enumerated
- 2026-08-11 · RK1078: a capture names the engine that failed and nothing compares it with the engine that reads it
- 2026-08-11 · RK1077: nothing holds the verb surface complete, so a state no command closes is found by the project that reaches it
- 2026-08-11 · RK1076: the gate is silent on a partial line whose entry names no half, the one state whose repair is two verbs
- 2026-08-11 · RK1075: an open line the ledger records as whole has no verb that closes it, and the guard denies the hand edit
- 2026-08-11 · RK1074: fourteen module addresses are spelled by hand across the suite and nothing holds them level
- 2026-08-11 · RK1073: a refusal decided under a package subdirectory named no module at all
- 2026-08-11 · RK1072: the kernel declares sixty-four names the format has no use for
- 2026-08-11 · RK1071: budget prices a field against a limit and never says where the limit came from
- 2026-08-11 · RK1070: the grammar fold reads codes out of a list it also rewrites
- 2026-08-11 · RK1069: the kernel is a boundary a test holds and not a place the code lives
- 2026-08-11 · RK1068: a template whose separator can occur inside a field parses back differently and nothing refuses it
- 2026-08-11 · RK1067: a refusal names the limit it enforced and not the line of config that set it
- 2026-08-11 · RK1066: each kind of reference is resolved by its own code rather than by one rule
- 2026-08-11 · RK1065: the format's mechanism and this backlog's vocabulary share one namespace
- 2026-08-11 · RK1064: roadkeep.toml declares the limits and the markers but not which fields exist or where they sit
- 2026-08-11 · RK1063: the line format is stated twice, once in render and once in the parser
- 2026-08-11 · RK1062: budget --tools counts the tool list and not the handshake beside it
- 2026-08-11 · RK1061: the tool budget makes every lint build the whole served surface
- 2026-08-11 · RK1060: one paragraph about counting is repeated on thirteen served properties
- 2026-08-11 · RK1059: the largest every-turn surface is measured and held by nothing
- 2026-08-11 · RK1058: the skill teaches the count as a permission to replace and never names the span
- 2026-08-11 · RK1057: the wrapped refusal still describes the count as authorising only a deletion
- 2026-08-11 · RK1056: record add still publishes a --lines flag that exists only to be refused
- 2026-08-11 · RK1055: the served why bound refuses on the client a span the server accepts
- 2026-08-11 · RK1054: the ship that completes a partial reached disk without validating the entry it wrote
- 2026-08-11 · RK1053: two ledger writes still delete the continuation lines RK1049 taught one to keep
- 2026-08-11 · RK1052: record amend walks every entry for the reversal clause and prints none
- 2026-08-11 · RK1051: an id other entries cite can never be given the entry it is missing
- 2026-08-11 · RK1050: record add is described by one motivating case, so the door is not found for the others
- 2026-08-11 · RK1049: record amend parses a multi-line entry and cannot write one back, so correcting one collapses it
- 2026-08-11 · RK1048: show says an id was never written when a ledger entry delivers it without leading with it
- 2026-08-11 · RK1047: body.promise names three doors and not reserved_ids, the mechanism built for the state it fires on
- 2026-08-11 · RK1046: a partial-marker line beside an unqualified entry has no door and no finding
- 2026-08-11 · RK1045: a refusal offers as its remedy the exact command that produced it
- 2026-08-11 · RK1044: retire refuses an id the ledger already holds without naming ship, the verb that does close the line
- 2026-08-11 · RK1043: the four slash commands spell a console script the plugin's other two surfaces were written not to assume
- 2026-08-11 · RK1042: delivered lists a reverted entry as a delivery, unmarked, and it is the read made before every add
- 2026-08-11 · RK1041: budget on a new task reports the assumed section's occupancy, so words-left describes a section the task will not use
- 2026-08-11 · RK1040: init writes the Non-goals heading and no [non_goals], so the verb that fills it refuses on a fresh project
- 2026-08-11 · RK1039: export answers a missing target with an errno and an absolute path, where the state next to it names the fix
- 2026-08-11 · RK1038: defer and retire call the prose field --reason where nine other verbs call it --why
- 2026-08-11 · RK1037: agents.md is at 125 of 125 lines, so the next module added to the package makes the gate red
- 2026-08-11 · RK1036: a section budget reports the limit that does not bind, so the aim promises room the write refuses
- 2026-08-11 · RK1035: the ancestor row names an add on a section that exists, and neither figure is what an amend may write
- 2026-08-11 · RK1034: the ancestor refusal tells an amend to pick a free top-level anchor, which is a door only an add has
- 2026-08-11 · RK1033: section amend accepts a body that puts its parent over, which is the write RK1024 and RK1029 left open
- 2026-08-11 · RK1032: an unambiguous prefix reaches --fix, --prune and --claim, so a two-letter typo is a write nobody asked for
- 2026-08-11 · RK1031: an id a project reserves for an epic and never writes as a line is reported as spent, on every prose mention, forever
- 2026-08-11 · RK1030: a roadkeep.toml a Windows editor saved with a BOM answers 'Invalid statement (at line 1, column 1)' and names no byte
- 2026-08-11 · RK1029: budget answers a whole limit free for a child whose parent has one word left, which is the read an add is composed from
- 2026-08-11 · RK1028: a body read from a path keeps its editor's BOM, and neither the write nor the gate reports the invisible byte
- 2026-08-11 · RK1027: The forward-reference guard has no advice for the case it most often fires on: an id that is about to exist
- 2026-08-11 · RK1026: An unrecognised flag falls through to argparse, so a near-miss gets the verb list and its own value echoed back
- 2026-08-11 · RK1025: A section address given to show is answered as an unknown task id, though section show is the verb for it
- 2026-08-11 · RK1024: add accepts a subsection that lint then fails, because the parent section's total is not checked at insert
- 2026-08-11 · RK1023: the --why - stdin escape hatch is unusable on PowerShell, which prefixes a BOM the char.invisible check rejects
- 2026-08-11 · RK1022: add refuses for a missing --ref and the anchors it sends you to names --family, which add has no flag for
- 2026-08-11 · RK1021: Nothing enumerates the questions a maintainer asks a governed file, so the law about answering them holds nothing
- 2026-08-11 · RK1020: Six names across four modules are imported and never spelled, and nothing in this tree reports one
- 2026-08-10 · RK1019: The harness proves the client against whatever roadkeep the machine resolves, and nothing says it was this checkout
- 2026-08-10 · RK1018: The backlog tree opens with no count, so how much work is left is a question only reading the file answers
- 2026-08-10 · RK1017: Every save re-asks which copy of the tool answered, so a keystroke costs a subprocess and a git call
- 2026-08-10 · RK1016: The Layout index names a sixth surface nowhere, and what holds it reads package modules only
- 2026-08-10 · RK1015: A door says nothing about whether running it writes, and the kind beside it describes the remedy and not the door
- 2026-08-10 · RK1014: A project with no queue is told to write the heading by hand, and the verb that would move one says the first writes it
- 2026-08-10 · RK1013: The host is a folder, so installing it means copying a directory and nothing here produces the file an editor takes
- 2026-08-10 · RK1012: A pointer can resolve to a heading with no prose, or to one with no title, and the gate reports neither
- 2026-08-10 · RK1011: No package this repository builds is one an editor can load, so nothing it answers is reachable without a shell
- 2026-08-10 · RK1010: A surface written in another language has no gate in this tree, so CI proves the Python and nothing proves the client
- 2026-08-10 · RK1009: Nothing says which roadkeep an editor calls, so a view answers from a copy the project does not write with
- 2026-08-10 · RK1008: Adding a task from the editor means retyping the fields in a terminal, so a limit is met after the sentence exists
- 2026-08-10 · RK1007: A violation is a line of terminal output, so the editor showing that file marks nothing at the column that failed
- 2026-08-10 · RK1006: The backlog is legible only in a terminal, so the editor holding the file offers the raw text and a hand edit
- 2026-08-10 · RK1005: No test reads a `--json` payload as an outside client would, so a renamed key breaks one and nothing here fails
- 2026-08-10 · RK1004: The register reads gate to write and nothing reads it back, so a rule only the door holds is found by writing one
- 2026-08-10 · RK1003: A design naming an id no line carries is refused at the door and reported by nothing once a file already holds one
- 2026-08-10 · RK1002: An id named only as an example in prose is read as spent, and nothing refuses it where the prose is composed
- 2026-08-10 · RK1001: A second defect in a class an invariant already holds reads like new work, so a gap in a holder is filed as a fix
- 2026-08-10 · RK1000: A limit, prefix or marker written into the package instead of read from config is found by an adopting project
- 2026-08-10 · RK500: A dep flag naming an id nothing carries is written, and so is one on the block the line is being filed into
- 2026-08-10 · RK499: Every prose door accepts a tab, a zero-width codepoint and a space that is not one, and the gate refuses all three
- 2026-08-10 · RK498: Nothing says which of the gate's findings a write can refuse, so a rule the write path drops is found by tripping it
- 2026-08-10 · RK497: A ship writes a why naming a path the repository lacks, and the gate then refuses the file the ship just wrote
- 2026-08-10 · RK496: A test surveying the package's source keeps passing once a directory of modules stops being covered by its glob
- 2026-08-10 · RK495: A scope declaring a directory stages every file under it and reports each one as claimed by nothing
- 2026-08-10 · RK494: 82 handlers sit in the file that registers them, so no verb family has a module or a docstring of its own
- 2026-08-10 · RK493: Every command's output rendering shares one file with the parser and 82 handlers, 8,489 lines
- 2026-08-10 · RK492: Nothing counts the sites an invariant has yet to reach, so a migration's remaining work is found by tripping over it
- 2026-08-10 · RK491: Three tests hold three surface-wide rules under no common name, so a rule nothing holds reads as one that is held
- 2026-08-10 · RK490: A remedy row repeats what its finding already carries, so a row naming the wrong file or verb is found by example
- 2026-08-10 · RK489: Each subparser checks its own subjects and narrowing flags, so a swallowed flag is found one verb at a time
- 2026-08-10 · RK488: Every module that offers a command composes the sentence itself, so a new surface is reached one emitter at a time
- 2026-08-10 · RK487: A pipe in a block title breaks the readme table, where the site form escapes the same string
- 2026-08-10 · RK486: The sweep holding the spelling class renders each message once, and two of the four have a shape it never sees
- 2026-08-10 · RK485: adopt says the prefix was inferred from the ids on a file where it read no id at all
- 2026-08-10 · RK484: The merge driver's capture line offers an argv naming three files git has already deleted
- 2026-08-10 · RK483: Two branches that each shipped a task cannot merge the prose file, though neither touched the other's section
- 2026-08-10 · RK482: A line one branch removed and the other edited is reported as both branches rewriting it
- 2026-08-10 · RK481: A capture the default report writes cannot be replayed, and the session that could still take one is not told
- 2026-08-10 · RK480: Five tasks moved a message off the shell and nothing holds the sixth, which is found by hand or not at all
- 2026-08-10 · RK479: The attestation half of the Stop names a shell lint, where the half beside it spells the session's own
- 2026-08-10 · RK478: The gate that blocks a turn lists its findings without the door RK420 gave each one
- 2026-08-10 · RK477: The deny offers both spellings for every write and only the shell one for the reads it closes on
- 2026-08-10 · RK476: The follow-up an add leaves is a shell argv, in the payload a tool-surface caller acts on
- 2026-08-10 · RK475: A refusal reaching an agent over the tool surface names the shell, on the write it just refused
- 2026-08-10 · RK474: The doors repair never dispatches are checked against nothing, so a printed one may refuse too
- 2026-08-10 · RK473: Nothing runs repair over a real corpus, so a remedy that cannot close its finding is found by hand
- 2026-08-10 · RK472: A remedy that refuses every time is re-attempted on every run, and nothing marks it as stuck
- 2026-08-10 · RK471: The count repair closes with says three ran where one ran and two are printed FAILED above it
- 2026-08-10 · RK470: A remedy naming a section omits which prose file it is in, so on a project with two it opens the wrong one
- 2026-08-10 · RK469: Twenty-six findings state one fact about two files, and one config line closes all of them
- 2026-08-10 · RK468: One finding names two different commands, and repair runs the one its own message does not
- 2026-08-10 · RK467: Nothing checks whether a flag pair answers as one of its halves, so each swallowed flag is found by hand
- 2026-08-10 · RK466: Two commands take two answers and print one, saying nothing about the flag that lost
- 2026-08-10 · RK465: A flag that narrows one budget subject is silently ignored beside three of the others
- 2026-08-10 · RK464: The tool list is 13,000 tokens a session and nothing budgets it, where an 8 KB resident file is gated
- 2026-08-10 · RK463: Eight reads the shipped skill tells an agent to call have no tool, on the install with no console script
- 2026-08-10 · RK462: A run naming several files gets no workers at all, so six files take fifteen seconds in series
- 2026-08-10 · RK461: Every row the ownership audit prints on this project is a standing memo, which is never a thing to act on
- 2026-08-10 · RK460: Running one test file is 33 times slower than before, because -n auto spawns a worker per core for it
- 2026-08-10 · RK459: The ownership an address states is reachable one family at a time, so an audit costs a call per family
- 2026-08-10 · RK458: A test passes alone and fails when another file's test has run before it in the same worker
- 2026-08-10 · RK457: The suite runs only in series, so a full run is five minutes before any change is known to hold
- 2026-08-10 · RK456: Eleven test files each build a git repository per test, with seven processes and a private copy of the helper
- 2026-08-10 · RK455: A CLI call whose stdin was already read dies while forcing UTF-8 instead of running the verb
- 2026-08-10 · RK454: A governed file with NUL bytes among its text is 400 findings whose named --fix changes nothing
- 2026-08-10 · RK453: No read says which outline sections a backlog still owns, so prose a departure left is found by grepping
- 2026-08-10 · RK452: Which of two writes came first decides whether an outline section ever belongs to a task
- 2026-08-10 · RK451: A governed file that is not text at all is reported one finding per byte, and never as a file to restore
- 2026-08-10 · RK450: A staged file is renamed into place without its bytes being flushed, so a crash leaves the new size and no content
- 2026-08-10 · RK449: A remedy reaches the tool surface as a shell argv, so the caller least able to run one is handed it
- 2026-08-09 · RK448: Two of the guard's four routes still name the shell where this session has the tool that answers
- 2026-08-09 · RK447: The deny lists tools this session has no server for, ahead of the shell commands it does have
- 2026-08-09 · RK446: An empty --near is answered with the whole block instead of being refused
- 2026-08-09 · RK445: adopt lists a block once per heading, so a ledger grouped by sub-headings reports the same label three times
- 2026-08-09 · RK444: An agent that never hand-edits is never told the tools exist, the deny being the only place that names them
- 2026-08-09 · RK443: A stored capture that never reached the verb replays as reproducing the symptom it was filed under
- 2026-08-09 · RK442: delivered prints a whole block's ledger to answer whether one proposal collides with it
- 2026-08-09 · RK441: Three copies price a lexical duplicate match at 33rd, and re-measuring this ledger does not reproduce it
- 2026-08-09 · RK440: A capture whose re-run stopped at a usage refusal is filed as reproducing the symptom above it
- 2026-08-09 · RK439: A sub-heading nested under a block's own heading is read as a second declaration of that block
- 2026-08-09 · RK438: A write's event line calls a block empty that every query in the same tool calls finished
- 2026-08-09 · RK437: Prose is filled to a width counted in code points while every limit that refuses counts UTF-16 units
- 2026-08-09 · RK436: The MCP schema publishes a limit that counts code points while the write refuses in UTF-16 units
- 2026-08-08 · RK435: The door offered for a queued block that has no lines yet says the block was never declared
- 2026-08-08 · RK434: The gate decides whether a block is finished, empty or undeclared with its own copy of the rule
- 2026-08-08 · RK433: delivered answers a block letter nothing declares as one that has shipped nothing
- 2026-08-08 · RK432: A dep on a block nobody has filed anything under resolves as satisfied
- 2026-08-08 · RK431: An id mentioned in prose is consumed by the next add, so the sentence that named it is now wrong
- 2026-08-08 · RK430: The line-length limit counts code points and the gate that enforces it counts UTF-16 units
- 2026-08-08 · RK429: A query against a finished block answers the same as one against a block that never existed
- 2026-08-08 · RK428: The totality check cannot see a code the gate builds by f-string
- 2026-08-08 · RK427: The queue defect lint reports in roadkeep.toml is refused by the only verb that repairs a queue
- 2026-08-08 · RK426: A call breaching two limits is refused twice, the second time for a limit the first refusal already measured
- 2026-08-08 · RK425: The repair block.repeated names sends the caller into a worse state than the one it reported
- 2026-08-08 · RK424: A denied write answers with every verb its file has, not the one the defect needs
- 2026-08-08 · RK423: A code is the only stable name a defect has and nothing will say what it means
- 2026-08-08 · RK422: Nothing turns a report into the commands that clear it
- 2026-08-08 · RK421: Nothing holds the tool to having a door for every defect it reports
- 2026-08-08 · RK420: 25 of the gate's 37 codes report a defect and name no verb for it
- 2026-08-08 · RK419: A project whose ledger is organised by nothing can no longer write its first line
- 2026-08-08 · RK418: Two copies at one version number are called agreed, on the one fact known not to tell trees apart
- 2026-08-08 · RK416: A line proposing what a revert already undid reads as new work
- 2026-08-08 · RK414: A slip of the pen in a symptom has to borrow the verb for a false premise
- 2026-08-08 · RK410: The free address is the last line of a listing of spent ones
- 2026-08-08 · RK409: The machine-readable brief answers a finished block in prose
- 2026-08-08 · RK408: The ship that empties a block reports it and names no verb for it
- 2026-08-08 · RK407: A refusal blames the author for a character its own shell wrote
- 2026-08-08 · RK402: The tree that ships the plugin is told to wire the guard a second time
- 2026-08-08 · RK401: A line the gate calls clean is refused by the door that corrects it
- 2026-08-08 · RK400: Shipping a family's last child leaves its parent describing deleted work
- 2026-08-08 · RK399: One field is spelled --status on add and --marker on resume
- 2026-08-08 · RK385: Nothing notices that a new line asks for what a shipped entry already delivered
- 2026-08-08 · RK366: The skill every adopting project loads has 24 body lines past 110 characters against a file otherwise wrapped at 90
- 2026-08-07 · RK417: The finding for one label under two headings names no verb, and one of the two is often removable
- 2026-08-07 · RK415: Three engines write and judge one project and none names the others
- 2026-08-07 · RK413: The first block heading lands after the section the blocks are supposed to precede
- 2026-08-07 · RK412: The seam every line write passes is told which file it is writing into twice
- 2026-08-07 · RK411: The add and the gate stay quiet about a ledger organised by nothing, and now a verb repairs it
- 2026-08-07 · RK406: The body flag that takes a path refuses the dash every other tool reads as standard input
- 2026-08-07 · RK405: A ledger organised by nothing is a project every ship refuses and no verb can repair
- 2026-08-07 · RK404: The refusal that sends an add to the block verb never says which file the line was going into
- 2026-08-07 · RK403: A block heading the changelog declares twice has no verb to fold into one, leaving only the hand-edit the guard denies
- 2026-08-07 · RK398: The version bump the hook writes is never staged again once both files are dirty
- 2026-08-07 · RK397: A prose paragraph whose line starts with a bold span is inserted verbatim instead of filled
- 2026-08-07 · RK396: A dep on a partially shipped task is annotated as satisfied
- 2026-08-07 · RK395: A shipped entry that gets reverted stays in the ledger saying it shipped
- 2026-08-07 · RK394: The wiring writes four surfaces and leaves the flag unhonoured when the driver's file cannot be written
- 2026-08-07 · RK393: The gate names a command that cannot run, on a tree the wiring half wrote and refuses to finish
- 2026-08-07 · RK392: A scaffold that cannot create a declared file's directory leaves the configuration behind, and both doors out refuse
- 2026-08-07 · RK391: The gate holds no one-heading-per-label rule, and the write path resolves a repeated one by position
- 2026-08-07 · RK390: A block label checked only to its first line, and never against the others, scaffolds a file no verb can address
- 2026-08-07 · RK389: A dep naming two real ids is accepted as one thing outside the backlog
- 2026-08-07 · RK388: A heading that parsed with its sigil is rewritten without one by a verb amending a different part of the file
- 2026-08-07 · RK387: The line telling an unread file from an empty one is on the backlog path only, so --sections still gives that zero
- 2026-08-07 · RK386: The oldest of the estimate's three argument refusals is the one no test holds
- 2026-08-07 · RK384: A prefix named with --sections is taken and never used, and a value the command refuses elsewhere passes silently
- 2026-08-07 · RK383: The free-address help says one outline spans both files, which a declared namespace makes false
- 2026-08-07 · RK381: A refusal on a short field costs the whole rationale a second time
- 2026-08-07 · RK380: A block can carry open lines for months and be found missing only by the first ship
- 2026-08-07 · RK379: The refusal for a missing anchor does not name the anchor to use
- 2026-08-07 · RK377: A doubled anchor has no verb that repairs it
- 2026-08-07 · RK368: A non-goal whose reason changes has no door but drop and re-add, which moves the bullet to the end of the list
- 2026-08-06 · RK376: A file that is not a backlog reports the headline an empty one gets, so a mistyped path reads as needing nothing
- 2026-08-06 · RK375: A pyproject that configures roadkeep measures as an empty backlog, the reading RK374 refused for the other declaration
- 2026-08-06 · RK374: The configuration measured as a backlog answers zero of everything, under a comment saying a later command refuses it
- 2026-08-06 · RK373: A prose file the run left out of its set is still absent from the unread list, so a check it never made reads as covered
- 2026-08-06 · RK372: One report calls a governed file both read and ungoverned, two functions resolving the same path by different rules
- 2026-08-06 · RK371: A doubling over files the caller named comes back in a field called `roles`, holding filenames no `[files]` answers
- 2026-08-06 · RK370: A file the estimate cannot open is refused in the OS's own sentence, with an absolute path and no argument named
- 2026-08-06 · RK369: A prose file the estimate is handed by name is read unqualified, so a namespace already declared reads as a collision
- 2026-08-06 · RK367: The two figures the no-size-field non-goal argues from are stated in five places and both have drifted
- 2026-08-06 · RK359: The doubled address the estimate reports is reachable only where the file is already one this project declares
- 2026-08-06 · RK358: The pick over this repository's own backlog is asserted lowest-ready, so a run made while a line is claimed is red
- 2026-08-05 · RK365: The function every gate check is threaded through grew a fifteenth branch, and what holds its size is an editor
- 2026-08-05 · RK364: The median this repository's own granularity claim is stated against has fallen below the floor a test holds
- 2026-08-05 · RK363: A mistyped address whose family exists is offered a new top-level instead of that family free child
- 2026-08-05 · RK362: The gate over near-twin verbs proves the sentence is declared and never that it reaches a caller
- 2026-08-05 · RK361: A line the merge driver re-places is refused without naming the file or the line it is about
- 2026-08-05 · RK360: A section written at an address no live task owns is refused with no way to learn a free one
- 2026-08-05 · RK357: A repair that removed a line reports the line number of whatever moved up into its place
- 2026-08-05 · RK356: Nothing says when the version this gate pins stopped being the one an installing user gets
- 2026-08-05 · RK355: What `--fix` repairs is enumerated in four places and none of them is derived from the fixer
- 2026-08-05 · RK354: The report names a file it does not list among the ones it checked, so a count and a finding disagree
- 2026-08-05 · RK353: A verb this tool publishes under one name on one surface is an unknown command on the other
- 2026-08-05 · RK352: The replay test that asserts a codec drifted names a value the running process may already declare
- 2026-08-05 · RK351: A test asserting on an MCP answer fails when any source file is touched while the suite runs
- 2026-08-05 · RK350: A third near-twin verb would be found by the session that types it, not by a test
- 2026-08-05 · RK349: The refusal that names the anchor command reaches one verb, and three others refuse the same way
- 2026-08-05 · RK348: A departure refused by a dependent's line reports the length and names no id
- 2026-08-05 · RK347: The estimate reads one prose file at a time, so the state two of them are in is the one it cannot report
- 2026-08-05 · RK346: One question is answered by two fields, and the older one names one namespace of several
- 2026-08-05 · RK345: The budget every other field is read before is the one an every-turn file has no door for
- 2026-08-05 · RK344: A capture embeds only the file a finding named, so a lint report from a three-file project cannot be replayed
- 2026-08-05 · RK343: A replay reports a defect as still reproducing from a staging holding none of the files the command reads
- 2026-08-05 · RK342: A claim's read-back reports the files its own transaction wrote as owned by nothing
- 2026-08-05 · RK341: A capture records no environment, so the field defect RK337 traced was unreplayable
- 2026-08-05 · RK340: outline anchors are one namespace across prose roles, and every outline starts at I
- 2026-08-05 · RK339: the verb that writes a marker is the one whose name reads as the report
- 2026-08-05 · RK338: The payload validator's own test cannot decode the report it asserts on, so it fails wherever the plugin is published
- 2026-08-05 · RK337: A crash in the environment's own stdio encoding is unreproducible and untested
- 2026-08-05 · RK335: The job that gates a release pipes an unpinned remote script into bash, under the workflow's default token
- 2026-08-05 · RK334: The payload validator skips wherever the Claude CLI is absent, which is every CI run that gates a release
- 2026-08-05 · RK333: The refusal offers mcp__roadkeep__add, which is not the name a plugin-provided server gives it
- 2026-08-05 · RK332: Nothing runs the payload's own validator, so a surface that loads with empty metadata passes CI
- 2026-08-05 · RK331: The skill's frontmatter is not valid YAML, so the loader drops its name and description in silence
- 2026-08-05 · RK330: Each feature needing state outside the documents invents its own file format and its own reader
- 2026-08-05 · RK329: the two prose arguments of one command disagree about stdin, and the shell-hostile one is the literal
- 2026-08-05 · RK328: The one queue repair that is fully derived has no fixer, so a clean tree still costs a hand edit
- 2026-08-05 · RK327: A departure leaves the priority entry that named it in place, so the queue outlives the line it ordered
- 2026-08-05 · RK326: The gate passes a priority entry naming a shipped id, an unknown id or a block nothing declares
- 2026-08-05 · RK325: The queue that outranks the id order lives in the one file nothing governs, so no verb keeps it true
- 2026-08-05 · RK324: brief calls a shipped task ready, so the one command meant to start work invites redoing it
- 2026-08-05 · RK323: The published payload carries this repository's own CLAUDE.md, which the loader warns is not context
- 2026-08-05 · RK322: The payload's root .mcp.json is auto-loaded as a plugin surface, naming a path no adopter has
- 2026-08-05 · RK321: The manifest declared the hooks file the loader already auto-loads, so no surface loaded at all
- 2026-08-05 · RK320: The version bump stages two whole files, so an unrelated edit to either rides into the next commit
- 2026-08-05 · RK319: A branch that refuses --json cannot be served at all, and nothing states the coupling
- 2026-08-05 · RK318: The command the staleness note tells the reader to run is not the act the tool was called for
- 2026-08-05 · RK317: The driver path accepts a --json it silently ignores, on the one path git actually invokes
- 2026-08-05 · RK316: Which bounds table describes a tool is a verb spelled in a function, where every other per-tool fact is a field
- 2026-08-05 · RK315: A test asserting on this repository's own docs fails when another session writes them mid-run
- 2026-08-05 · RK314: The marker argument publishes no enum on the one tool whose whole job is writing a marker
- 2026-08-05 · RK313: The one refusal a stale server does explain names only a remedy the agent cannot perform
- 2026-08-05 · RK312: The refusal that demands a ref names nothing that would produce one
- 2026-08-05 · RK311: A body refused for one word is discarded whole, so the check costs what the limit saves
- 2026-08-05 · RK310: A design section can carry a stale premise and ship drops it in silence
- 2026-08-05 · RK309: Every scope restates the governed files the tool itself just wrote
- 2026-08-05 · RK308: The verb that declares a scope has no tool on the surface it was built for
- 2026-08-05 · RK305: The scheme suggestion is suppressed by a majority that shipping erodes
- 2026-08-05 · RK304: The one closed-set argument on the MCP surface publishes no enum, so a role nobody declared is refused after the call
- 2026-08-05 · RK303: The section budget picks the first of two prose files declaring one anchor, where every other reader refuses
- 2026-08-05 · RK275: No MCP tool asks whether git would run this driver
- 2026-08-05 · RK267: A staleness note lists every changed module and hands relevance back
- 2026-08-04 · RK307: A scope corrected mid-task is restated in full or not at all
- 2026-08-04 · RK306: A closure leaves the claim its departure would have released
- 2026-08-04 · RK302: section add writes an anchor another declared prose file already holds, and only a retired one is refused
- 2026-08-04 · RK301: The pre-write budget is silent about the one field in the same transaction that also refuses
- 2026-08-04 · RK300: Whether a block still holds open work is computed in two modules
- 2026-08-04 · RK299: A report names every path relative to a root it never states
- 2026-08-04 · RK298: A shipped line's commit cannot read the scope its claim carried
- 2026-08-04 · RK297: The read that says which anchor address is free asks one prose file on a project with two
- 2026-08-04 · RK296: An undeclared-block refusal still lists every label the file declares
- 2026-08-04 · RK295: A scope naming a path the tree does not have is recorded in silence
- 2026-08-04 · RK294: Nothing calls the scope at the moment the commit is made
- 2026-08-04 · RK293: anchors names the next address under every family and not the next family
- 2026-08-04 · RK292: The scope line names files from the project that did not supply the target
- 2026-08-04 · RK291: Nothing in the estimate says which checks it did not run
- 2026-08-04 · RK290: The estimate cannot see any finding that needs more than one line
- 2026-08-04 · RK289: Only one payload is held to the dataclass it renders
- 2026-08-04 · RK288: A rationale file's anchors read as absent under the other scheme
- 2026-08-04 · RK287: A section's word count includes its subsections, so it names a figure no limit is measured against
- 2026-08-04 · RK286: The declaration a ledger reject names does not change the count
- 2026-08-04 · RK285: The estimate calls a conforming backlog a total rewrite
- 2026-08-04 · RK284: Uninstall reports keeping a workflow the project never had
- 2026-08-04 · RK283: budget answers for the task line only, so the two largest prose limits are discoverable only by failing
- 2026-08-04 · RK282: The refusal that most needs the adopt door is the one that does not name it
- 2026-08-04 · RK281: A rationale file not already using the sigil reads as no sections at all
- 2026-08-04 · RK280: Two sessions each ship a commit carrying the other's code
- 2026-08-04 · RK279: A backlog written as an ordinary Markdown list is priced at zero
- 2026-08-04 · RK278: The verb still names a driver the check has stopped asking for
- 2026-08-04 · RK277: The check demands a driver in a repository that routes nothing to it
- 2026-08-04 · RK276: One write reports a skip on only one of its two surfaces
- 2026-08-04 · RK274: The repair a check names takes over a wiring the same check just reported
- 2026-08-04 · RK273: Whether git sends a file to this driver is guessed from one file
- 2026-08-04 · RK272: The repair a driver check names writes one of the two halves it just failed on
- 2026-08-04 · RK271: The capture offer closes the gate's own answer as though the tool had failed
- 2026-08-04 · RK270: A driver check that answers current has not looked at .gitattributes
- 2026-08-04 · RK269: A block emptying is stated once to the console and recorded nowhere a later verb can read
- 2026-08-04 · RK268: A cached derivation survives the test that monkeypatched what it read
- 2026-08-04 · RK266: A registered driver that stopped resolving is silent until a merge
- 2026-08-04 · RK265: budget omits the pointer under an outline ref scheme, so it over-reports the room by the anchor width
- 2026-08-04 · RK264: The cost query spends 95% of its answer on records its own percentiles summarise
- 2026-08-04 · RK263: A test asserting about the live checkout fails the same way whether code or the tree moved
- 2026-08-04 · RK262: A section add writes a heading ship will later decline to delete
- 2026-08-04 · RK261: A denial still loads the document model and the lock for names it never reaches
- 2026-08-04 · RK260: Every denied edit imports the MCP server module to read a tuple of tool names
- 2026-08-04 · RK259: The body's word budget names a prose role the project may not declare
- 2026-08-04 · RK258: The three prose bodies publish no limit, so 250 words reaches the author as a refusal
- 2026-08-04 · RK257: A ship refusal names the missing block heading but not the verb that writes it
- 2026-08-04 · RK256: Messages outside the guard still spell the console script, the capture offer included
- 2026-08-04 · RK255: The merge driver git is told to run is a bare console script
- 2026-08-04 · RK254: The guard prints a shell command a plugin install does not provide
- 2026-08-04 · RK253: A field this project closed is refused as one that does not exist
- 2026-08-04 · RK252: The declaration a closed field names is written beside its predicate, not with it
- 2026-08-04 · RK251: A conditional argument with no reason to give raises while composing the refusal
- 2026-08-04 · RK250: The drift note asserts a wiring it never observed
- 2026-08-04 · RK249: The add answer names the id it derived and not the pointer
- 2026-08-04 · RK248: The drift note reads the launch directory, not the root the wiring is a fact about
- 2026-08-04 · RK247: Nothing says which outline anchors history still cites
- 2026-08-04 · RK246: The staleness note names a reload that cannot fire on the tree it fires in
- 2026-08-04 · RK245: The number that binds an amend is the one figure `budget` states only in characters
- 2026-08-04 · RK244: A paused id cannot be named as what supersedes a retired line
- 2026-08-04 · RK243: The one refusal no prose edit fixes now advises deleting words
- 2026-08-04 · RK242: The stale-import notice makes a correct refusal look like a bug
- 2026-08-04 · RK241: The MCP add tool cannot write a line on an outline-ref project
- 2026-08-04 · RK240: Where an id actually is gets composed by two modules in the same words
- 2026-08-04 · RK239: Two prose files declaring one anchor is silent until a line points at it
- 2026-08-04 · RK238: `anchor.unknown` names the one remedy RK4 closes for an id the ledger holds
- 2026-08-04 · RK237: A block whose heading carries a pre-adoption note can never be withdrawn
- 2026-08-04 · RK236: A departing line takes a standing memo's subsection with it
- 2026-08-04 · RK233: A non-goal `brief` prints by its lead is one `drop` cannot address
- 2026-08-04 · RK232: Writer and gate part company again on an anchor two prose files declare
- 2026-08-04 · RK231: A paused task's rationale cannot be corrected, the door refusing it as prose nothing owns
- 2026-08-04 · RK230: `add --section` cannot write into the only prose file a project declares
- 2026-08-04 · RK229: `defer` reports the carried design as kept in a file it never asked declares it
- 2026-08-04 · RK201: A length refusal states its surplus in the one unit the author composing the retry cannot count
- 2026-08-04 · RK200: Which governed files no verb wrote is answerable only by trying to end the turn
- 2026-08-03 · RK235: `install` reads the checkout that ships the plugin as a project to wire into
- 2026-08-03 · RK234: Nothing in an adopting project runs the gate that holds its vendored skill in step
- 2026-08-03 · RK228: How git spells a token is recomputed from scratch per token, and twice for most of them
- 2026-08-03 · RK227: The invisible-codepoint check calls a Python predicate once per character of every governed file
- 2026-08-03 · RK226: A path written with Windows separators is a path on one platform and a filename on the other
- 2026-08-03 · RK225: A run judging a revision stats the working tree once per token and throws every answer away
- 2026-08-03 · RK223: The path check reads every ledger entry twice, and the gate takes seconds on a large ledger
- 2026-08-03 · RK222: show pays two git listings on every call, on the read that starts every task
- 2026-08-03 · RK221: show and lint disagree about whether a deleted artefact is a path the task names
- 2026-08-03 · RK220: One path that escapes the repository disables the ignore check for every other path in the ledger
- 2026-08-03 · RK219: A cache on the tree answers whichever question it was asked first, and its key is nothing
- 2026-08-03 · RK218: The same revision answers two different things about a path depending on an untracked working file
- 2026-08-03 · RK217: Deleting a whole tracked directory silences the check that exists to notice a deleted artefact
- 2026-08-03 · RK216: A refusal named a block the caller never mentioned, and it is a prefix of theirs
- 2026-08-03 · RK215: A top-level section's own prose cannot be corrected once it has subsections
- 2026-08-03 · RK214: A ledger that declares no marker closes one of the three doors out of the roadmap
- 2026-08-03 · RK213: A gate whose answer turns on whether an ignored build directory happens to be on this machine
- 2026-08-03 · RK212: A citation of a shipped design cannot be told from a citation of one that never existed
- 2026-08-03 · RK211: Every governed write now re-derives two projections and nothing measured what that costs
- 2026-08-03 · RK210: A pinned corpus config reports six artefacts missing from a tree it deliberately does not carry
- 2026-08-03 · RK209: section drop deletes prose other sections cite and reports nothing, the door the ship fix did not reach
- 2026-08-03 · RK208: A block only the ledger declares made every projection raise instead of counting it as finished
- 2026-08-03 · RK207: The screen's copy of the guard's path rule was held by nothing, so widening one would open a hole in the other
- 2026-08-03 · RK206: A section other prose cites is deleted by ship, and the gate reads only the pointer a task line carries
- 2026-08-03 · RK205: The package publishes annotations that no checker is allowed to read
- 2026-08-03 · RK204: A denied Edit pays the whole package import that a shell command no longer does
- 2026-08-03 · RK203: The layout index grows with the package and the budget holding it cannot tell it from prose
- 2026-08-03 · RK202: The CLI parser is a pure function of the code and is rebuilt for every message
- 2026-08-03 · RK199: Every entry point pays a schema import to reach a module that does not read one
- 2026-08-03 · RK198: Every tool call rebuilds the whole CLI parser twice, and the argument check is what pays
- 2026-08-03 · RK197: `add` asks for a rationale section that another declared prose file already holds
- 2026-08-03 · RK196: A pointer into the strategy file survives the ship that deletes the line pointing at it
- 2026-08-03 · RK195: The roadmap's own amend would strand the same tail, and nobody has counted whether a line wraps there
- 2026-08-03 · RK194: No query shows the lines a wrapped entry owns, so what a correction replaces is read off the file
- 2026-08-03 · RK193: Completing a partial rewrites the first line of an entry whose sentence runs past it
- 2026-08-03 · RK192: A pinned corpus hands out a config whose file reads go to the live tree instead
- 2026-08-03 · RK191: The refusal on a second partial names the entry in the way, not the id the step needs
- 2026-08-03 · RK190: A line's prose budget is discoverable only by writing a sentence and being refused
- 2026-08-03 · RK189: A path claim is satisfied by a file of that name anywhere in the tree, and the floor was never measured
- 2026-08-03 · RK188: Every write stales the derived README block, so the gate fails until somebody remembers `export`
- 2026-08-03 · RK187: `export --readme --site` writes the first file and then refuses the second, leaving one refreshed
- 2026-08-03 · RK186: `brief` reports a pointer into the strategy file as resolving to nothing
- 2026-08-03 · RK185: The budget an author is given is in characters, the one unit a model cannot count
- 2026-08-03 · RK184: A length refusal names the rendered total and not the field that is over or the surplus
- 2026-08-03 · RK183: The two field limits sum to the line limit, so a line obeying both is refused by the third
- 2026-08-03 · RK182: The round-trip corpus reads two foreign roadmaps and no foreign ledger, where the wrapped shape lives
- 2026-08-03 · RK181: A section refusal names the file by its whole filesystem path, where the gate beside it names it relatively
- 2026-08-03 · RK180: A subsection is written at a fixed depth, so in a file whose top level is deeper it is no subsection
- 2026-08-03 · RK179: Correcting a wrapped entry's why leaves the rest of the old sentence below the new one
- 2026-08-03 · RK178: A line whose bold symptom turns out to be false has no repair
- 2026-08-03 · RK177: The tool list can change mid-session and the client is never told, so it validates against a cached schema
- 2026-08-03 · RK176: Widening the matcher to Bash spawns an interpreter per shell command and nothing measured it
- 2026-08-03 · RK175: An approved shell write leaves a valid line and no record of who wrote it
- 2026-08-03 · RK174: Listing the tools rebuilds the whole CLI parser fifty-two times and costs a sixth of a second
- 2026-08-03 · RK173: A path a ledger entry names relative to its own module is reported as not in the repository
- 2026-08-03 · RK172: A pointer resolving to a section in the strategy file is reported as resolving to nothing
- 2026-08-03 · RK171: Nothing asks which exposed tool can reach a read of the transport
- 2026-08-03 · RK170: A tool call that reads stdin eats the transport it arrived on
- 2026-08-03 · RK169: section drop deletes rationale a live line points at, when the anchor is a bullet
- 2026-08-03 · RK168: A busy checkout makes the gate exit like a violation, so a clean file reads as a failing one
- 2026-08-03 · RK167: Three read-only commands take the write lock themselves, each in a different place
- 2026-08-03 · RK166: A newly declared block cannot receive its first rationale section
- 2026-08-03 · RK165: Clearing a stale row without a marker write means deleting every live claim with it
- 2026-08-03 · RK164: A claim on a paused or shipped id is reported as though no line ever carried it
- 2026-08-03 · RK163: A release drops one id and reconciles nothing, so every row no door reported stays in the listing
- 2026-08-03 · RK162: A shipped line leaves an entry in the registry that nothing will ever read again
- 2026-08-03 · RK161: Which lines are claimed is answerable only as a side effect of asking for work
- 2026-08-03 · RK160: Setting the in-progress marker on a line another worker holds takes the claim without a word
- 2026-08-03 · RK159: Two places write the claim on the one path that takes one, and the second overwrites the first
- 2026-08-03 · RK158: Marking a line in progress by hand asserts nothing, so the next caller is sent at it
- 2026-08-03 · RK157: A new ledger entry lands between the last entry and its own wrapped lines
- 2026-08-03 · RK155: A tool call answers with the code the session started, so a refusal can name a key the config declares
- 2026-08-03 · RK148: install wires four surfaces and leaves the merge driver to whoever remembers it
- 2026-08-03 · RK147: A prose file's own limits are enforced by the gate and ignored by the write that creates the text
- 2026-08-03 · RK146: A tab is reported as an invisible character and repaired by nothing, for ever
- 2026-08-03 · RK145: A block can only be opened last, and where it sits is what every ordered answer reads
- 2026-08-03 · RK144: Declaring a block writes three files and nothing takes the heading back out
- 2026-08-03 · RK143: A ledger entry filed under the wrong block can be corrected by no verb at all
- 2026-08-03 · RK140: The CI workflow `install` writes is red on its first run for every backlog with debt
- 2026-08-03 · RK139: `adopt` measures the task lines and says nothing about the non-goals it will govern
- 2026-08-03 · RK138: Un-wiring a project is the hand-edit this tool exists to deny
- 2026-08-03 · RK137: The copied skill names a shell command that does not exist in the project it was copied into
- 2026-08-03 · RK136: The section budget charges a measured table exactly what it charges a paragraph
- 2026-08-03 · RK135: A rationale section whose task points at a different one is reported by nothing
- 2026-08-03 · RK134: A section that four open lines point at is reported stale, and the drop it names is refused
- 2026-08-03 · RK132: The README block is read and written back with nothing checking the file did not move
- 2026-08-03 · RK131: The check that a transaction's files have not moved is taken before writes that take it again
- 2026-08-03 · RK128: The guard denies Edit and Write and answers silence to a shell command writing the same file
- 2026-08-03 · RK125: Declaring an unmarked ledger takes a whole verb away, and the refusal offers no way through
- 2026-08-03 · RK122: id.two-files calls a correct half-shipped state a contradiction
- 2026-08-03 · RK114: A subsection whose task is gone is exempt from the ownership check that would report it
- 2026-08-03 · RK111: The MCP write path cannot write an id the deriver never mints
- 2026-08-03 · RK110: `adopt` counts the id findings without saying they are one declaration
- 2026-08-03 · RK107: A project that declares the format still has nothing enforcing it
- 2026-08-03 · RK105: A concurrent edit in another repository turns this project's suite red
- 2026-08-03 · RK104: Nothing gates the README block this tool writes, so a stale restatement passes lint
- 2026-08-03 · RK103: A bullet whose marker holds a space is read as prose and reported by nothing
- 2026-08-02 · RK156: Renaming an id drops the claim on that line, so work somebody is holding is offered again
- 2026-08-02 · RK154: The one call a session starts a task with never names the lines a claim kept out of its answer
- 2026-08-02 · RK153: A checkout whose plugin version never moves is one the session keeps running the old copy of
- 2026-08-02 · RK152: A started task that is blocked is reported to every caller with nothing saying somebody holds it
- 2026-08-02 · RK151: Every project's claims expire after the same hour, and how long a task takes is not the same everywhere
- 2026-08-02 · RK150: The query an agent runs most may now ask permission, for a flag it did not pass
- 2026-08-02 · RK149: The door the skill sends an agent through to start work cannot take the line it hands over
- 2026-08-02 · RK142: The ledger inherits the roadmap's problem statement by default, and no verb can revise it
- 2026-08-02 · RK141: Opening a block needs a heading only a hand-edit can write, and the guard denies that edit
- 2026-08-02 · RK130: A transaction that stopped after its ledger write can be finished by no command
- 2026-08-02 · RK129: Retiring a task whose half already shipped deletes the record that it did
- 2026-08-02 · RK127: record drop assumes a duplicate is redundant, and picks the later entry either way
- 2026-08-02 · RK126: Corruption inside a ledger entry is reported by lint and repairable by no verb
- 2026-08-02 · RK124: A ledger entry can be written and deleted but never corrected
- 2026-08-02 · RK123: An open task's rationale cannot be corrected by any verb, and the hook denies the hand-edit
- 2026-08-02 · RK121: A task delivered in halves has no representation, so the corpus invents one the grammar cannot read
- 2026-08-02 · RK120: Two branches each spend the same id, and the merge git produces is a conflict in a file only this tool may write
- 2026-08-02 · RK119: Every agent asking what to work on is handed the same line, including one another agent already started
- 2026-08-02 · RK118: A reader can load a governed file truncated, and an interrupted ship leaves two of its three files written
- 2026-08-02 · RK117: Two adds started together mint the same id, and the gate reports it only once both lines exist
- 2026-08-02 · RK116: A second write to a governed file erases the first one's line and both commands exit 0
- 2026-08-02 · RK115: Where a heading's region ends is re-derived at every call site, under two rules that already differ
- 2026-08-02 · RK113: renumber renames a section's heading and leaves the anchors of everything nested under it
- 2026-08-02 · RK112: section drop deletes a section an open line points at
- 2026-08-02 · RK109: The id's shape is declared once and read by two parsers that answer differently
- 2026-08-02 · RK108: The first line added to a block lands above the paragraph that introduces it
- 2026-08-02 · RK106: An id a live backlog spells the same way every time is refused as malformed
- 2026-08-02 · RK102: A test over a live corpus fails when the other project ships the work it counted
- 2026-08-02 · RK101: An outline anchored by block letter declares no section, and one by block C declares one
- 2026-08-02 · RK100: A project that runs roadkeep from a checkout has no write-path skill
- 2026-08-02 · RK99: Nothing measures the rationale file an adopting project brings
- 2026-08-02 · RK98: A table-shaped backlog reads as zero lines, which is what an empty file reads as
- 2026-08-02 · RK94: weight gives one commit's whole size to every task it shipped, so a bulk import pins the percentiles to itself
- 2026-08-02 · RK92: The dep resolver has no answer for a dep blocked on paused work
- 2026-08-02 · RK84: lint cannot answer whether this change made it worse
- 2026-08-02 · RK83: pick offers an idea when a designed task is ready
- 2026-08-02 · RK21: A standard adopted by one project is a preference
- 2026-08-01 · RK97: an id two branches allocated to different work has no door that moves either line
- 2026-08-01 · RK96: There is no state for a task set aside but neither shipped nor abandoned
- 2026-08-01 · RK95: an id no commit ever carried is reported as history being unreadable
- 2026-08-01 · RK93: add exits 0 on a line whose derived pointer fails lint at once, and names no follow-up
- 2026-08-01 · RK91: Pausing a task means retiring it, losing its id, rationale section and every dependent that named it
- 2026-08-01 · RK89: A capture survives the command that produced it and not the session that ran it
- 2026-08-01 · RK78: ship deletes the sections nested under the one it drops
- 2026-07-31 · RK90: A command declared to tolerate a broken config still crashed on a TOML syntax error
- 2026-07-31 · RK88: A field report cannot be told fixed from unfixed without restaging the reporter's project
- 2026-07-31 · RK87: A capture taken inside a private repository is a disclosure the moment it is filed
- 2026-07-31 · RK86: A refusal names the rule it enforced and never what to do when the rule is the defect
- 2026-07-31 · RK85: A defect an agent hits in another repository is reported as prose, or is not reported at all
- 2026-07-31 · RK82: Nothing announces the write path at the moment a session starts reading the files
- 2026-07-31 · RK81: The MCP server is declared and no tool reaches the client
- 2026-07-31 · RK79: The plugin cache and the developer checkout are two engines, and both answer 0.1.0
- 2026-07-31 · RK76: The estimate a project takes before adopting is not the one the gate applies
- 2026-07-31 · RK75: A backlog that files work under any other word cannot declare its headings
- 2026-07-31 · RK74: A backlog numbered by track cannot declare its ids
- 2026-07-31 · RK73: A non-goal could be written and never corrected, so a wrong lead was the Edit the hook denies
- 2026-07-31 · RK72: The non-goal against estimates names a schedule, not the field a size bucket would be
- 2026-07-31 · RK71: A task that should be two looks exactly like one that should be one
- 2026-07-31 · RK70: No command can write a non-goal, and the denial names five that cannot
- 2026-07-31 · RK69: Non-goals are printed when a task starts, not when one is proposed
- 2026-07-31 · RK68: The brief prints a non-goal narrower than the file states
- 2026-07-31 · RK67: A ledger entry written twice cannot be removed, so the file states one decision twice
- 2026-07-31 · RK66: A project whose convention is no section means no pointer cannot declare it
- 2026-07-30 · RK65: A line can be created and removed but never corrected, so every adoption defect needs a hand-edit
- 2026-07-30 · RK64: Shipping one of several tasks that share a section deletes the design the others still point at
- 2026-07-30 · RK63: A heuristic read off this repository disabled the new check on the corpus it was written for
- 2026-07-30 · RK62: A roadmap line for work already in the ledger has no door out of the roadmap
- 2026-07-30 · RK61: Under the outline scheme no check can say which task a section belongs to
- 2026-07-30 · RK60: A project that adopts the tool with drift already in it has its every turn blocked by history
- 2026-07-30 · RK59: Four tools cover four commands, so the eight a task actually needs still require a shell
- 2026-07-30 · RK58: A refusal names a command the machine may not have, now that the plugin needs no install
- 2026-07-30 · RK57: Installing the plugin is not enough: every surface it declares calls a console script the user has to install first
- 2026-07-30 · RK55: A MIME type, an i18n key and a list of method names are reported as missing files
- 2026-07-30 · RK54: Deleting a section one line at a time validates half-deleted states nobody wrote
- 2026-07-30 · RK53: An example line inside a fenced code block is read as a task and counted
- 2026-07-30 · RK52: A shipped line is told to move its second sentence to a section that was deleted on ship
- 2026-07-30 · RK51: A link written relative to its own file is read as a path the repository lacks
- 2026-07-30 · RK50: One budget judges a roadmap line and a ledger of history nobody will rewrite
- 2026-07-30 · RK49: A follow-up nested under its parent line is invisible to every count and every pick
- 2026-07-30 · RK48: A ledger whose lines carry no marker reads as prose, and every dep on one goes unknown
- 2026-07-30 · RK26: A plugin installed by hand is a plugin one project has
- 2026-07-30 · RK25: A human driving the same standard should not have to learn the CLI
- 2026-07-30 · RK24: Shelling out puts argument names in prose, where they are guessed
- 2026-07-30 · RK23: Rules resident every turn spend the budget they exist to protect
- 2026-07-29 · RK47: A lettered outline subsection is no anchor, so 20 of Turing's headings are sections nothing sees
- 2026-07-29 · RK46: All 8 paths reported missing on Shio are globs, placeholders or files the task will create
- 2026-07-29 · RK45: A section belonging to no task lands after the last block, where it reads as that block's rationale
- 2026-07-29 · RK44: Under ref_scheme = outline a heading numbers itself, so 151 headings yield zero sections
- 2026-07-29 · RK43: A changelog whose lines carry no status marker parses as zero entries and zero rejects
- 2026-07-29 · RK42: The landing page claimed 9 of 36 tasks shipped through twenty-five ships
- 2026-07-29 · RK41: A fix that was never planned has no door into the ledger
- 2026-07-29 · RK40: A block's own next task is not askable, so a global answer reads as that block being finished
- 2026-07-29 · RK39: A README and a site restate a backlog they cannot re-read, so both are stale from the first ship
- 2026-07-29 · RK38: A write reports an exit code and nothing about what it changed, so only a human can react to it
- 2026-07-29 · RK37: A dep on a block that does not exist resolves as satisfied
- 2026-07-29 · RK36: A rationale section can gain a requirement the line carrying its status never mentions
- 2026-07-29 · RK35: A dep on a range or a block hides how much work it actually names
- 2026-07-29 · RK34: An invisible character reports a visible error about something else
- 2026-07-29 · RK32: An id that exists in neither file was retired for a reason nobody recorded
- 2026-07-29 · RK31: The reasoning behind a shipped decision survives only in a commit nobody greps
- 2026-07-29 · RK30: The instruction file loaded every turn has a budget nothing enforces
- 2026-07-29 · RK29: Starting a task costs reading two whole files to learn one line's worth of context
- 2026-07-29 · RK28: A dep the tool cannot resolve reads exactly like one that is merely pending
- 2026-07-29 · RK27: A pointer's target is numbered by hand, and deleting a section leaves a hole nobody renumbers
- 2026-07-29 · RK22: An agent can hand-edit the file the CLI is supposed to own
- 2026-07-29 · RK20: Shio's 92 active lines average 142 words against a one-sentence rule
- 2026-07-29 · RK19: Installing from a git clone keeps a standard local
- 2026-07-29 · RK18: A tool that requires an empty repo cannot be adopted by the repo that needs it
- 2026-07-29 · RK17: A gate that runs only on a developer's machine is not a gate
- 2026-07-29 · RK16: A report of ninety-two violations is a report nobody acts on
- 2026-07-29 · RK15: A pointer to a section that does not exist reads as a design that does
- 2026-07-29 · RK14: A format that is documented but not enforced is a format that drifts
- 2026-07-29 · RK13: A blocked task looks identical to a ready one, and one that unblocks half the backlog to one that unblocks nothing
- 2026-07-29 · RK12: A task's design lives in a second file and nothing joins them
- 2026-07-29 · RK11: Picking work means reading the whole file to find one task whose deps are shipped
- 2026-07-29 · RK10: Counting a backlog by grep silently drops the lines it fails to match
- 2026-07-29 · RK9: The four files are not four of the same thing, and prose has no line to validate
- 2026-07-29 · RK8: A dep annotation goes stale the moment its target ships
- 2026-07-29 · RK7: Two files can disagree about one task's status
- 2026-07-29 · RK6: Shipping a task is four edits across three files, so one is always missed
- 2026-07-29 · RK5: Writing the line by hand is where the prose leaks in
- 2026-07-29 · RK4: The next id cannot be inferred from a block header, and a wrong guess collides with a retired one
- 2026-07-29 · RK3: Hardcoding one project's vocabulary makes the tool single-use
- 2026-07-29 · RK2: A parser that cannot re-render what it read corrupts the file it edits
- 2026-07-29 · RK1: Nothing knows what a task line is, so every check is a regex over prose
- undated · RK1697: roadkeep-gui is a second repository, so a CLI change reaches its CI only through a separate checkout of this one