Check claims and verify work

Use evidence proportionate to the claim. A routine statement may need a direct read or test; selected work may need agreed executable acceptance criteria and an independent reviewer. Neither route introduces a universal workflow.

Install

/plugin marketplace add JRichlen/agent-plugins
/plugin install verify-before-claim@jrichlen

Choose the procedure

• Check a claim: inspect the source, run the relevant check and state what remains uncertain.
• Verify work: create and pin an acceptance contract, protect it while work proceeds, then obtain independent review.

Core reporting and authority rules remain binding without selecting a procedure. The method returns to the existing task record; it creates no coordinator, mandatory planning round, retry policy, model tier or automatic observer.

The shell helpers require Bash and coreutils. Claude Code can additionally use the contract-write hook. Other hosts need supported equivalent enforcement; script integrity/freshness checks do not establish that an independent reviewer actually ran. Preserve that distinction in the handback.

Caller migration guidance describes the retired product identity and thin source-path adapters. Earlier claim-check measurements are historical findings, not validation of this integration or proof of benefit across hosts.

PORTABILITY: On another harness without the Claude Code write hook, use a supported equivalent write guard or check pinned hashes before and after work. Hash checks detect drift afterward; they do not prevent writes. If prevention is required and unavailable, report that unmet prerequisite rather than claiming protected execution. Independent review still requires an actual reviewer.