reviewlisted
Install: claude install-skill compforge/devloop
devloop review has one lifecycle:
```text
ReviewRun → Finding → Verdict → Handling → thread resolution
```
`ccr:label=` is only the Forge encoding of a Verdict. Never format it manually and never use
generic `pr reply` for adjudication; the review CLI validates the vocabulary and keeps recording a
Verdict separate from resolving the discussion.
`<PLUGIN_ROOT>` maps to `${CLAUDE_PLUGIN_ROOT}` on Claude Code and `${PLUGIN_ROOT}` on Codex.
## Routes
### Surfaced ReviewRun result
When context contains `Review: ...`, treat it as advisory information, not control transfer:
- `running` → leave it alone;
- `stale` → report that the background run likely died and can be rerun;
- `clean` → no action;
- failed files / `error` → report incomplete coverage; never call it clean;
- findings → inspect the real diff/code before reporting them.
Use the Verdict vocabulary below instead of inventing a High/Medium/Low truth model. By default,
report the result briefly and return control. Modify code only when the user asks to fix or process
the findings.
```bash
<PLUGIN_ROOT>/scripts/python <PLUGIN_ROOT>/scripts/review.py status
<PLUGIN_ROOT>/scripts/python <PLUGIN_ROOT>/scripts/review.py run [--background '<context>']
```
Normal lifecycle hooks launch the same ReviewRun automatically; do not duplicate a visible
`running` run.
### Published Findings awaiting Verdicts
1. **List published Findings**:
```bash
<PLUGIN_ROOT>/scripts/python <PLUGIN_ROOT>/scripts/review.py findings <n