Claude Code First, specification version 0.1.0. The plain-language guide
Version 0.1.0. Level profile. Best evidence grade R3/S4.
AP-1 A ceiling on what the surface costs
Each surface a product serves to agents, its tool schemas, skill descriptions and bodies, and command responses, MUST carry a token ceiling that a gate enforces, raised only with its reason written beside the number or in the commit that raises it.
| Field | Value |
|---|---|
| Keyword | MUST |
| Level | profile |
| Checked | by judgement |
| Findings | F406, F4 |
| Harness facts | H3, H9 |
| Threat | internal: freewilly's budget borrows Shio's argument |
| Status | active |
Rationale: Shio, roadkeep, freewilly and winwright each measured what their surface costs a consuming agent and held it under a ceiling read by a test or lint (F406, R3/S4). Shio's tool-list test failed on a real change, "3 tokens over its 2200 ceiling" [shio@821f18d74:docs/CHANGELOG.md#L22], and a description was trimmed instead. The ceiling did not stop growth in roadkeep, where the session ceiling was raised twelve times; what it did was make each raise a written decision with the task that paid for it (F4, R3/S4), which is why the reason is part of the rule. The two costs are priced apart because a skill's description is paid every session and its body only when used (H3). H9 bears on tool schemas: with tool search on, the documented default, not every definition is sent up front, so a schema ceiling measures serialized size, which is not in doubt, rather than a cost every session is certain to pay. A change to H9 re-opens that part.