review-loop

# Review Loop Drive a code/PR change through independent review, verification, disposition, repair, and full re-review. Each finding ends in a verified repair, a cited drop, a successor handover, or declared residual. Measure the artifact against the project's own stated goal and governing conventions, in its declared authority order. ## Caller Signature ``` /review-loop [source?] [scope?] [landing?] source : codex | code-review scope : PR number | implicit current-branch PR / working tree landing : head | stacked (PR scope only) ``` Read designations from the request's words as well as its arguments. `source` selects the reviewer; the host is the environment driving the loop, not another user choice. Read only that host's reference before determining availability: | Host | `source=code-review` | `source=codex` | Reference | |---|---|---|---| | Claude Code | Fork the available Claude `/code-review` skill | Spawn `codex exec` | [Claude Code](references/host-claude-code.md) | | Codex | Spawn `claude -p` with the available `/code-review` skill | Spawn a fresh `codex exec` | [Codex](references/host-codex.md) | Other hosts may supply the same capabilities. Record the actual host, reviewer, and execution route; a new process provides a separate review context, not evidence of an independent model family. Sources remain runtime parameters rather than fixed frontmatter dependencies. ## Source Interface ``` (diff pointer, design intent) → { findings[], verdict, exerc...