Agent guidance: landed contributor docs, review skill and CLI handoff

Updated 11 September 2026. The entry points now have separate, concrete owners.

| Entry point | State |
| --- | --- |
| `AGENTS.md`, `STATEMENTS.md`, `PROOFS.md`, `CONTRIBUTING.md` | Landed contributor guidance |
| Semantic review procedure and retained report contract | #4899, open for review |
| CLI usage guide, skill discovery/installation and prepare → finish handoff | #5386, draft after #4899/#5375 |

The contributor's existing agent conducts semantic review and owns authentication/model choice. The CLI prepares exact inputs, builds independently, collects bounded sources and validates the completed report. Findings are advisory; maintainers decide acceptance.

The review skill points to deterministic operations rather than reproducing them in prose. A separate plugin framework, model launcher, autonomous `formalize`/`fix` system and automatic broad review are deferred in #4394.

**Evaluation:** the earlier development scores in this discussion are historical and are not a broad accuracy estimate. The retained five-case pilot received William's qualitative approval; two original coverage gaps remain. No per-case scores or missed-issue estimates were supplied. See #4899 and #5376 for current evidence and limitations.

Close when the review and CLI entry points are accepted and available to contributors. Preserve the useful rubric/skills discussion below; it does not imply approval to enable an upstream review bot.

Read the landed guidance in AGENTS.md, STATEMENTS.md, PROOFS.md, and CONTRIBUTING.md, then follow the review work in #4899 and the CLI handoff work in #5386. Confirm that the review and CLI entry points are accepted and available to contributors; preserve the rubric discussion without enabling an upstream review bot.

Written by the indexing model from the issue text.