karlorz/agent-skills/claude-plan-status-review

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Review and verify Claude plan statuses (active/follow_up/review_needed) under ~/.claude/plans by cross-checking GitHub PR state and the latest repository code, then update plan metadata and generate a prioritized status-review report. Use when active/follow_up plans may already be implemented, blocked, or obsolete and you want a recurring review loop.

Install

npx vskill@latest install karlorz/agent-skills/claude-plan-status-review

SKILL.md

Claude Plan Status Review

Quick start

  1. Generate a status-review report:

``bash python3 scripts/status_review.py --out /Users/karlchow/.claude/plans/PLAN_STATUS_REVIEW.md ``

Notes:

  • The report includes Frontmatter Fixups (Copy-Paste) for review_needed plans; tune with --max-fixups.
  • If you already wrote a scan snapshot (recommended for scheduled runs), reuse it: --scan ~/.claude/plans/PLAN_SCAN.json.
  • claude-planctl runs (scan/archive) can take a couple minutes with GitHub checks; avoid short timeouts.
  • archive_candidate is not the same as “eligible to archive” (check eligibleForArchive / gate:* reasons in the report).

The report includes:

  • Plan title + any YAML frontmatter (repo, status, pr, updated, owner)
  • Any in-body Status: ... line (common in older plans)
  • Checkbox counts
  • Basic codebase verification: extracts referenced repo paths (e.g. apps/...) and checks if they exist on origin/HEAD for the local repo checkout (default ~/Desktop/code/<repo>).
  • A full Plan Index section (not truncated) for keyword/topic search across all plans (title + plan path + a few extracted path tokens like packages/<token>).
  1. Triage the report (oldest first) and update plans you touched:
  • Add/update frontmatter: repo, status, pr, updated, owner
  • Promote/demote: active ↔ follow_up as blockers change
  • Mark implemented only when you can point to merged code/PR
  1. If items are implemented, hand off archiving to claude-planctl (dry-run first):

``bash claude-planctl archive --dry-run --cooldown-days 7 --min-score 85 ``

Workflow

1) Build a review queue

Use claude-planctl scan as the source of truth for classification and GitHub state:

bash
claude-planctl scan --format table

Prioritize:

  • Oldest active and follow_up
  • Any review_needed
  • Any plan where a linked PR is closed/merged but the plan is still active/follow_up
2) Verify against latest codebase

For each plan you review:

  1. Identify repo + PR:
  • Prefer frontmatter repo / pr
  • Else use claude-planctl record fields (repo, prNumber, prUrl, primaryPrSource)
  1. Validate GitHub state:
  • Prefer claude-planctl GitHub check results
  • When needed: gh pr view <N> --repo <owner/repo>
  1. Validate against current code:
  • cd into the repo and git fetch --all --prune
  • Search for the plan’s key files/symbols (rg) to confirm whether work exists in main
  1. Decide status:
  • active: still relevant + actionable now
  • follow_up: blocked/deferred; capture the blocker in the plan body
  • implemented: clearly done (usually merged PR/code on default branch)
  • review_needed: ambiguous; do not guess
3) Update plan metadata (gradual adoption)

If a plan already has YAML frontmatter, update/add fields:

  • repo: owner/repo
  • status: active | follow_up | implemented
  • pr: 123 (when known)
  • updated: YYYY-MM-DD
  • owner: karlchow

If frontmatter is missing, add it only when you are already editing the plan for a real review.

Template:

yaml
---
repo: owner/repo
status: active | follow_up | implemented
pr: 123
updated: 2026-02-18
owner: karlchow
---
4) Archive implemented plans (conservative)

Defer to claude-planctl for archive moves (dry-run first, keep cooldown 7 days):

bash
claude-planctl archive --dry-run --cooldown-days 7 --min-score 85
claude-planctl archive --apply --cooldown-days 7 --min-score 85

Tip: if claude-planctl archive says “No eligible archive candidates”, check the report for gate:cooldown_not_met and re-run after the cooldown window.

For archive gate logic and status signal definitions, consult:

  • /Users/karlchow/.codex/skills/claude-plan-ops/references/status-signals.md
  • references/review-rubric.md

Outputs

  • scripts/status_review.py writes a prioritized report to PLAN_STATUS_REVIEW.md.

Useful flags:

  • --max-items 10 to keep the report short
  • --max-fixups 10 to reduce copy-paste frontmatter suggestions
  • --limit 200 to reduce scan time
  • --index-max-keywords 3 to reduce Plan Index keyword noise per plan
  • --repo-dir owner/repo=/abs/path to override local repo checkout paths
  • --no-github-check to skip GitHub API checks (faster, but less accurate PR state)
  • --no-git-fetch to skip git fetch (faster, less “latest-codebase” accurate)

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/karlorz/agent-skills/claude-plan-status-review/badge)](https://verified-skill.com/skills/karlorz/agent-skills/claude-plan-status-review)

Something wrong with this skill or its page? Tell us.