plan-reviewlisted
Install: claude install-skill gitgitWi/council-flow
# flow:plan-review — Multi-LLM plan critique
Plan-review is cheap insurance. Each model reads the same plan and tasks list independently, raises issues in its own voice, and the orchestrator (Claude) reconciles them into a Korean summary the user reviews. If the review surfaces meaningful gaps, the plan gets versioned (`artifacts/plan.v1.md`) and a new `plan.md` is written.
## Preconditions
- `<worktree>/.planning/<date>-<task>/plan.md` exists.
- `<worktree>/.planning/<date>-<task>/tasks.md` exists.
- Optionally `research.md`.
If any of these are missing, do not run plan-review — go back to `flow:plan` first.
## Reviewers
Use three reviewers by default; the diversity is the point. Model IDs come from `../../references/models.md` — do not hardcode here, and re-check that file when models move.
Default trio (diverse stack: codex + gemini + opencode):
- `artifacts/plan-review-codex.md` — `gpt-5.5` via `codex`
- `artifacts/plan-review-gemini.md` — `gemini-3.1-pro-preview` via `gemini`
- `artifacts/plan-review-kimi.md` — `opencode-go/kimi-k2.6` via `opencode`
Optional reviewers (ask the user before dispatching — see "Pre-flight + reviewer selection" below):
- `artifacts/plan-review-deepseek.md` — `opencode-go/deepseek-v4-pro` via `opencode` (deepest analysis, +10-15 min)
- `artifacts/plan-review-glm.md` — `opencode-go/glm-5.1` via `opencode` (fast extra opinion, +5-7 min)
If the user wants only two reviewers (token / time budget), keep Gemini + one of the Codex/OpenCode