Plan Review
Run the PLAN checkpoint gate that assembles reviewers from plan scope, dispatches agents in parallel, and blocks Build until every verdict is APPROVE.
Overview
plan-review is an agent skill most often used in Ship review (also Validate scope, Build pm) that auto-assembles a review panel and collects APPROVE verdicts on plan files before Build starts.
Install
npx skills add https://github.com/camacho/ai-skills --skill plan-reviewWhat is this skill?
- PLAN checkpoint: MUST receive APPROVE from all reviewers before proceeding to Build
- Auto-finds latest non-.done.md plan under ai-workspace/plans/ with --include/--exclude panel overrides
- Invokes assemble-panel from file types and keywords; technical-editor cannot be excluded
- Dispatches selected agents in parallel in a single message; fallback panel [technical-editor, code-reviewer] gate P2 cap
- Errors when no plan exists or plan is already .done.md finalized
- Fallback panel: technical-editor and code-reviewer with gate P2 and cap 3
- technical-editor cannot be excluded from the panel
Adoption & trust: 642 installs on skills.sh; 1 GitHub stars; 3/3 security scanners passed (skills.sh audits).
What problem does it solve?
You have an implementation plan but no deterministic, parallel reviewer gate—and you risk coding before stakeholders and quality agents sign off.
Who is it for?
Solo builders using ai-workspace/plans with Claude Code or Cursor who want a hard gate between planning and coding.
Skip if: Repos without ai-workspace/plans, one-off chat plans with no file artifact, or teams that skip formal plan review.
When should I use this skill?
PLAN checkpoint: plan file exists in ai-workspace/plans/ and you are ready for parallel reviewer verdicts before Build.
What do I get? / Deliverables
A scoped reviewer panel runs in parallel on the plan file; unanimous APPROVE clears the PLAN checkpoint so you can invoke build/implementation skills next.
- Logged reviewer panel with selection rationale
- Aggregated APPROVE/REJECT verdicts blocking or clearing Build
Recommended Skills
Journey fit
Spans multiple journey phases - primary shelf plus alternate fits below.
Canonical shelf is Ship review because the skill is an explicit phase gate between approved planning artifacts and implementation. Review subphase matches deterministic panel assembly, parallel agent dispatch, and verdict collection against ai-workspace/plans.
Where it fits
Lock scope in a plan file, then run plan-review so reviewers challenge assumptions before you commit calendar to Build.
Execute the PLAN checkpoint with deterministic panel selection from files-to-modify and plan keywords.
Hold implementation agents until plan-review returns unanimous APPROVE on the active plan.md.
Re-open review when operational fixes require a new non-.done plan before agent-driven patches land.
How it compares
Structured PLAN gate with parallel agents—not a single passive code-reviewer prompt in chat.
Common Questions / FAQ
Who is plan-review for?
Indie developers and small teams running camacho-style AI workspaces who need enforced plan sign-off before agents touch production code.
When should I use plan-review?
After Validate scope when a plan.md exists in Ship review; before Build pm or backend implementation; re-run when scope changes materially.
Is plan-review safe to install?
It orchestrates git ls and parallel agent reads of plan files—review permissions and the Security Audits panel on this Prism page before install.
SKILL.md
READMESKILL.md - Plan Review
Phase gate: PLAN checkpoint. MUST get APPROVE from all reviewers before proceeding to Build. ## Inputs - Plan file path (default: most recent non-`.done.md` file in `ai-workspace/plans/`) - `--include <agent>`: Force-add an agent to the panel - `--exclude <agent>`: Remove an agent (except technical-editor, which cannot be excluded) ## Steps 1. **Find the plan file**: ```bash ls -t "$(git rev-parse --show-toplevel)/ai-workspace/plans/"*.md | grep -v '.done.md' | head -1 ``` If no plan file found → error: "No plan file in ai-workspace/plans/. Write a plan first." If plan file is `.done.md` → error: "This plan is already finalized." 2. **Read the plan** and extract: - Files to Modify section - Full body text - Plan scope (count files changed, count body lines) 3. **Assemble the review panel** — invoke `/assemble-panel` with scope = plan file. It returns the panel based on file types and keywords in the plan. Pass `--include`/`--exclude` as overrides. Log which agents were selected and why. Codex/Cursor: read `.claude/skills/assemble-panel/SKILL.md` directly and apply the scope/keyword maps inline. If `/assemble-panel` is unavailable, fall back to `[technical-editor, code-reviewer]` with gate=P2, cap=3. 4. **Dispatch all selected agents IN PARALLEL** using the Agent tool (single message, multiple Agent tool calls). Each agent receives: - The plan file content - Their specific review mandate - Instructions to return: APPROVE / REVISE / DROP with findings tagged P0/P1/P2 5. **Collect verdicts** and synthesize into a summary table: ``` ## Plan Review Summary | Agent | Verdict | P0 | P1 | P2 | Key Finding | |-------|---------|----|----|-----|-------------| ``` 6. **Gate logic and review loop** — invoke `/assemble-panel` for panel governance. It provides the policy algebra (gate=P2, cap=3) and handles RETAIN, EXPAND, CONVERGE, and ESCALATE_RECURRING. Do not duplicate those rules here. After each round: - All APPROVE with no above-gate findings → "Proceed to Build" - Above-gate findings remain → fix, then re-review per assemble-panel's RETAIN + EXPAND rules - Cap reached with unresolved findings → escalate to human If `/assemble-panel` is unavailable, fall back: keep all reviewers with above-gate findings, exit at round 3 or when all clean. ## Panel Bounds - Minimum: technical-editor alone (1 agent, trivial plans) - Typical: technical-editor + code-reviewer + architect-reviewer (3) - Maximum: all agents (rare, large cross-cutting plans)