agentleFS
egregore-labsegregore-labs/egregore/.claude/skills/review-pr/SKILL.md

review-pr

Skill293 starsChanged 11 days agoMITView on GitHub

Use when the user says 'review PR', 'is this PR safe to merge', or 'audit PR' — runs a CTO-level 10-point review checklist on one or more pull requests. Not creating a PR (/pr) or validating local changes (/test).

  • Reads credentials

185 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use diagnose-ci-failures, review-delta and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • zerx-labSkill

    diagnose-ci-failures

    Diagnose CI failures for a PR using the GitHub CLI, extract error logs, and generate a plan to fix them. Use when the user asks to check CI status, pull CI issues, triage test failures, or investigate PR build failures.

    Also in 3 projects that use this

  • n24q02mSkill

    review-delta

    Review only changes since last commit using impact analysis. Token-efficient delta review with automatic blast-radius detection.

    Also in 3 projects that use this

  • GeiserXSkill

    docs-loop

    Use when the user asks to audit docs, check documentation for stale claims, or verify docs against the code.

    Also in 3 projects that use this

  • GeiserXSkill

    full-loop

    Drive a repository to done with sergio-loop, then keep improving it with refine-loop at a descending theta. Use only when the user explicitly invokes full-loop.

    Also in 3 projects that use this

  • n24q02mSkill

    build-graph

    Build or update the code review knowledge graph. Run this first to initialize, or let hooks keep it updated automatically.

    Also in 3 projects that use this

  • +7 more from GeiserX

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. 0 are common, 8 are distinctive, and 8 are very specific to this project.

  • 0

    Common

    Most of the 178 edited versions we read keep these.

  • 8

    Distinctive

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Not this: "/pr" (create a PR), "/test" (validate local changes)

    line 13
  2. Distinctive:

    CRITICAL: Suppress raw output: Never show raw JSON, raw diffs, or unformatted gh output.

    line 17
  3. Very specific:

    Auth: Run every gh command through bash bin/gh-run.sh.

    line 21
  4. Distinctive:

    Read the configured repository name:

    line 36
  5. Very specific:

    Use each printed value as {github_org} or {repo_name} below.

    line 42
  6. Very specific:

    Use {pr_number} from the command's arguments or the selected PR metadata, and {github_org} and {repo_name} from configuration:

    line 47
  7. Very specific:

    Use {github_org} and {repo_name} from configuration with the credential helper:

    line 53
  8. Distinctive:

    Check that all imports reference functions/modules that exist

    line 73
  9. Very specific:

    Check that Cypher syntax is valid (balanced parentheses, proper MATCH/RETURN/WHERE structure)

    line 74
  10. Very specific:

    Check that shell scripts use correct flags for the target platform (e.g.

    line 75
  11. Very specific:

    Check that referenced files/scripts actually exist in the codebase (e.g.

    line 76
  12. Very specific:

    Do hooks always exit 0 (observe/report hooks must never block)?

    line 134
  13. Distinctive:

    Wait for all delegated or inline reviews to complete.

    line 179
  14. Distinctive:

    If multiple PRs touch the same files:

    line 192
  15. Distinctive:

    After showing the summary, offer:

    line 271
  16. Distinctive:

    After merge, update local develop:

    line 285

“Common” is measured: kept by at least 6 in 10 of the 178 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Review a pull request with CTO-level scrutiny. Designed for vibe-coded PRs.

Arguments: $ARGUMENTS

When to invoke

User says: "review PR", "check this PR", "review #123", "is this PR safe to merge", "audit PR", "review cem's PRs" Not this: "/pr" (create a PR), "/test" (validate local changes)

Execution rules

CRITICAL: Suppress raw output. Never show raw JSON, raw diffs, or unformatted gh output. All output should be structured assessment.

For diff review, delegate to a subagent when available and permitted in this session; otherwise do it inline. When delegation is available and permitted in this session, each PR gets its own background subagent for parallel analysis. In either case, fetch the diff, analyze it, and return a structured verdict.

Auth: Run every gh command through bash bin/gh-run.sh; the helper reads and trims GITHUB_TOKEN from the project's .env and passes it directly to the child environment when nonempty, otherwise preserving the existing login. Never read, print, or type the secret.

Step 0: Determine scope

Parse $ARGUMENTS:

  • PR number (e.g. 123, #123) → review that single PR
  • Author (e.g. alice, bob) → list all open PRs from that author
  • --all → list all open PRs
  • Empty → list all open PRs targeting develop

Get org/repo:

bash bin/config-get.sh github_org

Read the configured repository name:

bash bin/config-get.sh repo_name

Use each printed value as {github_org} or {repo_name} below; quote it in single quotes when you use it in a command, and write any single quote inside the value as '\''.

Step 1: Fetch PR metadata

Single PR

Use {pr_number} from the command's arguments or the selected PR metadata, and {github_org} and {repo_name} from configuration:

bash bin/gh-run.sh pr view '{pr_number}' --json title,author,additions,deletions,changedFiles,headRefName,body,createdAt,state --repo '{github_org}/{repo_name}'

Multiple PRs (author or --all)

Use {github_org} and {repo_name} from configuration with the credential helper:

bash bin/gh-run.sh pr list --state open --json number,title,author,additions,deletions,changedFiles,headRefName,createdAt --limit 50 --repo '{github_org}/{repo_name}'

Filter by author login if specified.

Step 2: For each PR, run the 10-point checklist

Delegate each PR to a subagent when available and permitted in this session, using general-purpose subagents and running reviews in parallel when available and permitted in this session; otherwise review each PR inline. Use this same review context for each PR:

  1. The PR number
  2. The repo
  3. The requirement to invoke gh through bash bin/gh-run.sh from this checkout
  4. The full checklist below

The 10-Point CTO Checklist

Each review must fetch the diff through bash bin/gh-run.sh pr diff '{pr_number}', using the selected PR number, and evaluate:

1. Does it actually work?

  • Check that all imports reference functions/modules that exist
  • Check that Cypher syntax is valid (balanced parentheses, proper MATCH/RETURN/WHERE structure)
  • Check that shell scripts use correct flags for the target platform (e.g. base64 -d vs -D on macOS)
  • Check that referenced files/scripts actually exist in the codebase (e.g. bin/graph-batch.sh, bin/graph-op.sh)

For an Egregore framework PR, inspect affected skill instructions as part of this check. Use an isolated checkout at the PR's exact head SHA with its base ref available; do not switch the user's working branch or mix their unsaved changes into the review. Run the existing development engine against that checkout, using {pr_checkout} for its absolute path and {pr_base_ref} for that PR's available base ref:

bash bin/node-run.sh bin/capability-distribution.mjs skill-references \
  --root '{pr_checkout}' --changed --base '{pr_base_ref}' --branch-only --review

The receipt names consuming skills and their maintained instruction paths, including native runtime resources. Open the relevant instructions and check whether they still describe the changed behavior. This is direct-reference coverage, not proof of transitive dependency completeness. Report a missing base or unavailable engine as a coverage gap. The receipt is advisory and adds no approval checkpoint; include actual findings in the existing verdict.

2. Side effects on shared state

  • Does it modify .claude/settings.json? (hook changes affect every session)
  • Does it modify CLAUDE.md? (behavioral changes affect every session)
  • Does it modify bin/graph-op.sh? (new operations must not break existing ones)
  • Does it modify bin/session-start.sh? (startup changes affect every session)
  • Does it modify .env, egregore.json, or auth flows?

3. Overlap detection

  • Are there other open PRs touching the same files?
  • Is this PR a subset of a larger mega-PR? (common with vibe coding — person keeps working, creates multiple PRs from evolving branch)
  • Would merging this and another PR cause conflicts?

4. Destructive operations

  • Graph: Any DELETE, DETACH DELETE that could remove data?
  • Files: Any file deletions that could break other commands? Check if deleted files are referenced elsewhere.
  • Endpoints: Any API endpoint removals that could break the website or other consumers?
  • Schema: Any node/relationship type removals?

5. Security regression

  • Does it remove boundary rules, permission checks, or security guards?
  • Does it remove .syncignore entries (controls what gets synced to public repo)?
  • Does it weaken auth flows or token validation?
  • Does it expose secrets (hardcoded tokens, API keys, credentials)?
  • Does it add world-readable temp files with sensitive content?

6. Schema consistency

  • New node types or relationships must be declared in CLAUDE.md schema line
  • Property naming must follow existing conventions (camelCase, not snake_case)
  • Relationships must use UPPER_SNAKE_CASE

7. Idempotency

  • Graph writes must use MERGE not CREATE for edges/nodes that could be created twice
  • Scripts that run on schedule or can be retried must produce the same result on re-run
  • Exception: CREATE is OK for truly unique entities (e.g. new Session per session)

8. Error handling in critical path

  • Does set -euo pipefail at script top risk killing the script before essential output?
  • Are graph queries guarded with || true or fallback values?
  • Do hooks always exit 0 (observe/report hooks must never block)?

9. Convention violations

  • .env must never be sourced (source .env breaks on spaces) — use grep '^KEY=' .env | cut -d'=' -f2-
  • Hardcoded absolute paths from developer's machine (e.g. /Users/cemdagdelen/...)
  • Committed lock files, .DS_Store, or session-specific state
  • Dead code (unreachable cases, unused imports)

10. Stale base

  • How far behind develop is the branch?
  • Are there guaranteed merge conflicts with current develop?
  • Has the same file been modified on develop since the PR was created?

Agent output format

Each review must return a structured assessment:

PR: #NNN — Title
Author: name
Branch: branch-name
Size: +X/-Y, N files
Created: date

CHECKLIST:
1. Works:          PASS/FAIL/WARN — details
2. Side effects:   PASS/FAIL/WARN — details
3. Overlap:        PASS/FAIL/WARN — details
4. Destructive:    PASS/FAIL/WARN — details
5. Security:       PASS/FAIL/WARN — details
6. Schema:         PASS/FAIL/WARN — details
7. Idempotency:    PASS/FAIL/WARN — details
8. Error handling: PASS/FAIL/WARN — details
9. Conventions:    PASS/FAIL/WARN — details
10. Stale base:    PASS/FAIL/WARN — details

BLOCKING ISSUES:
- [list or "None"]

RISK: LOW / MEDIUM / HIGH
VERDICT: MERGE / FIX THEN MERGE / NEEDS WORK / REJECT

Step 3: Compile results

Wait for all delegated or inline reviews to complete. Sort PRs into tiers:

Tier classification

TierCriteriaAction
Tier 1: Easy merge0 FAILs, 0-2 WARNs, LOW riskMerge immediately
Tier 2: Fix then merge1-2 FAILs (minor), MEDIUM riskFix specific issues, then merge
Tier 3: Needs work3+ FAILs or HIGH riskSend back with fix list
Tier 4: RejectSecurity regression, broken imports, or fundamentally wrong approachClose with explanation

Overlap resolution

If multiple PRs touch the same files:

  1. Identify which is the "superset" (mega-PR)
  2. Recommend merge order to minimize conflicts
  3. Flag PRs that should be closed as superseded

Step 4: Render summary

Single PR review

┌──────────────────────────────────────────────────────────────────────┐
│  ✧ REVIEW                                         oz · Mar 11      │
├──────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  PR #241 — Fix session date sort                                     │
│  Author: cem · Branch: dev/cem/identity-fragmentation-fix            │
│  Size: +11/-11 · 4 files · Created: Mar 6                           │
│                                                                      │
│  CHECKLIST                                                           │
│    ✓ Works              Correct Cypher pattern                       │
│    ✓ Side effects       API files only                               │
│    ✓ Overlap            None                                         │
│    ✓ Destructive        No deletes                                   │
│    ✓ Security           No regression                                │
│    ✓ Schema             No changes                                   │
│    ✓ Idempotency        N/A (read queries)                           │
│    ✓ Error handling     Existing guards preserved                    │
│    ✓ Conventions        Clean                                        │
│    ⚠ Stale base         5 days old, low conflict risk                │
│                                                                      │
│  BLOCKING: None                                                      │
│                                                                      │
├──────────────────────────────────────────────────────────────────────┤
│  Risk: LOW · Verdict: MERGE                                         │
└──────────────────────────────────────────────────────────────────────┘

Multi-PR review (sorted by tier)

┌──────────────────────────────────────────────────────────────────────┐
│  ✧ REVIEW                                         oz · Mar 11      │
│  14 open PRs from cem · 10 unique (4 superseded)                    │
├──────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  TIER 1 — EASY MERGE                                                │
│    #241  Fix session date sort              +11/-11    LOW    MERGE  │
│    #267  Handoff claiming                   +58/-2     LOW    MERGE  │
│    #185  Rewrite /meeting                   +586/-734  LOW    MERGE  │
│                                                                      │
│  TIER 2 — FIX THEN MERGE                                            │
│    #266  PR linkage                         +167/-7    MED    FIX    │
│          → Cypher bug: LIMIT before SKIP                             │
│    #265  /summon command                    +207/-0    LOW    FIX    │
│          → Missing schema declaration                                │
│                                                                      │
│  TIER 3 — NEEDS WORK                                                │
│    #244  Deep-reflect redesign              +604/-711  MED-HI WORK  │
│          → Non-idempotent edges, bash YAML parsing                   │
│    #269  Granola API upgrade                +1096/-228 MED    WORK  │
│          → Bundled unrelated changes, User-Agent spoof               │
│                                                                      │
│  TIER 4 — REJECT / CLOSE                                            │
│    #202  Extract CLAUDE.md into skills      +299/-247  HIGH   REJCT │
│          → Security rules deleted, not extracted                     │
│    #268  Coder workspace cleanup            +179/-488  HIGH   REJCT │
│          → .syncignore deleted, missing import, role overwrite       │
│                                                                      │
│  SUPERSEDED (close)                                                  │
│    #264  → by #256                                                   │
│    #265  → by #256                                                   │
│    #272  → by #256                                                   │
│                                                                      │
├──────────────────────────────────────────────────────────────────────┤
│  Merge order: #241 → #267 → #185 → #266 → #265                     │
│  3 ready · 2 fixable · 2 need work · 2 reject · 3 superseded        │
└──────────────────────────────────────────────────────────────────────┘

Step 5: Offer next actions

After showing the summary, offer:

  • "Start merging Tier 1?" — if there are easy merges
  • "Show details for #NNN?" — deep dive on a specific PR
  • "Fix #NNN?" — checkout the branch, apply fixes, push

Merge execution (when user confirms)

For each PR the user confirmed for merging, use its metadata number as {pr_number} and the configured {github_org} and {repo_name}, and merge through the gate:

bash bin/agent.sh merge 'https://github.com/{github_org}/{repo_name}/pull/{pr_number}'

When the gate refuses on a red check, show the user the failing jobs it names, and merge only if they decide those jobs do not matter: write a receipt naming each one and why, then run the same command with --receipt <file>, and the gate posts it on the PR before it merges.

After merge, update local develop:

git fetch origin develop --quiet

For PRs confirmed for closure as superseded, use their metadata number as {pr_number}, the superseding PR number as {superseding_pr_number}, and the configured {github_org} and {repo_name}:

bash bin/gh-run.sh pr close '{pr_number}' --comment 'Superseded by #{superseding_pr_number}' --repo '{github_org}/{repo_name}'

Edge cases

ScenarioHandling
PR already mergedSkip, note in output
PR has merge conflictsFlag in stale-base check, cannot auto-merge
PR targets main (not develop)Warning — all PRs should target develop
Author is maintainer (oz)Still run full checklist, but note trusted author
Mega-PR detectedIdentify sub-PRs, recommend close superseded
No open PRs"No open PRs found."
Graph offlineSkip Cypher validation checks, note limitation

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.