agentleFS
akshaypimprikarakshaypimprikar/pragma/.claude/skills/pipeline-review/SKILL.md

pipeline-review

Skill4 starsChanged 6 days agoMITView on GitHub

Audit the agent pipeline itself for drift, staleness, gaps, and inefficiencies, producing a severity-rated report. Runs automatically after every release and on a weekly schedule; can also be invoked manually.

13 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run this entire audit in the background.

    line 16
  2. Very specific:

    Check every file in .claude/skills/ for references to skills that are not in the current registry.

    line 21
  3. Very specific:

    If this project was derived from a workflow template (e.g.

    line 29
  4. Distinctive:

    If over budget, list the specific sections that could be trimmed.

    line 37
  5. Distinctive:

    Check whether the following gates exist as skill files or documented steps in the pipeline:

    line 48
  6. Very specific:

    After saving the report, use the PushNotification tool:

    line 103
  7. Very specific:

    If there are zero findings, body: Pipeline is healthy — no issues found.

    line 107

“Common” is measured: kept by at least 6 in 10 of the 13 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

Pipeline Review Agent

You are the Pipeline Review Agent for an iOS app project. Your job is to audit the agent pipeline for drift, staleness, gaps, and inefficiencies — and produce a severity-rated, actionable report.

Trigger

Runs automatically after every /release and on a weekly schedule. Can also be invoked manually: /pipeline-review.

Process

Run this entire audit in the background. Save findings and send a push notification when done.

What to audit

1. Stale skill references

Check every file in .claude/skills/ for references to skills that are not in the current registry.

Current valid skills: plan, spec, design, review, feature, test, bugfix, release, gates, pipeline-review, sync-workflow, trim-context, simplify, security-review, code-review, code-review:code-review, ios-build-verify, ios-coverage, ios-swiftdata-test-fixture, update-config, keybindings-help, fewer-permission-prompts, schedule, loop, claude-api, init, claude-code-setup:claude-automation-recommender, run, verify, status, benchmark, parallel-review, pr-followup, pragma-review, deterministic-pr-gates

Flag any skill name used in a skill or command file that does not appear on this list. Severity: Critical.

2. Template drift

If this project was derived from a workflow template (e.g. pragma), compare .claude/skills/ against the template. Flag:

  • App-specific content in the template that should use placeholders — High
  • Skill files that exist in this project but have no template equivalent — Medium
  • Logic improvements in this project's skills not yet back-ported to the template — Low
  • Logic improvements present in the template but not yet pulled into this project's skills — Medium (higher than its counterpart above: this direction means the live pipeline is running stale/buggy logic already fixed elsewhere, not just a future scaffold missing an enhancement; /sync-workflow only pushes project → template, so this comparison is the only thing that catches it)

3. AGENTS.md token budget

Count lines in AGENTS.md (or CLAUDE.md, for a project still on the pre-AGENTS.md convention). Target: ≤50 lines. If over budget, list the specific sections that could be trimmed. Severity: Medium if 51–60 lines, High if >60 lines.

4. Memory staleness

Read every file in the memory directory for the directory Claude Code was launched from: ~/.claude/projects/<launch-dir path with every non-alphanumeric character replaced by ->/memory/ (the sed in trim-context/SKILL.md computes it). If you launch from a parent folder that holds several projects, that is the parent's directory, not this project's (which may exist but be empty). Flag any memory that:

  • References a branch, PR, or task that no longer exists in git log or gh pr list
  • Describes a failure mode that has since been fixed in the skill files (the fix is the source of truth)
  • Has not been confirmed valid in >60 days and makes a specific factual claim Severity: Medium per stale entry.

5. Pipeline gate coverage

Check whether the following gates exist as skill files or documented steps in the pipeline:

GateRequired locationSeverity if missing
Pre-PR gate (/gates).claude/skills/gates/SKILL.mdCritical
Build verification (separate from tests)Gate 1 of /gatesHigh
CHANGELOG incremental updateStep in /featureHigh
/test before /gates; code-review before /review (app lane only)feature/SKILL.md Done-when + pr-followup/SKILL.md + AGENTS.md/CLAUDE.mdHigh
Lane config and required checks (gates, review-evidence)scripts/pipeline_lanes.json, .github/workflows/gates.yml, .github/workflows/review-evidence.ymlCritical
Coverage gate (Gate 6)scripts/check_gates.shMedium
Security check for sensitive PRssecurity-review skill referenceMedium
Session recovery command (/status).claude/skills/status/SKILL.mdLow

6. Skill file completeness

Each skill or command file must have: Trigger, Process, Done when sections. Flag any file missing a section. Severity: Medium.

7. Settings hygiene

Read .claude/settings.json. Flag:

  • Hooks referencing non-existent files or scripts — High
  • Missing UserPromptSubmit hook for unaddressed pipeline-review alert — Medium
  • Permissions that are overly broad where tighter patterns would suffice — Low

Output format

Save to docs/pipeline-review/YYYY-MM-DD.md with this exact structure:

---
date: YYYY-MM-DD
addressed: false
---

# Pipeline Review — YYYY-MM-DD

## Critical
- [ ] <finding> — `<file>:<line>` — **Fix:** <one-line recommended action>

## High
- [ ] <finding> — **Fix:** <recommended action>

## Medium
- [ ] <finding> — **Fix:** <recommended action>

## Low
- [ ] <finding> — **Fix:** <recommended action>

## All clear
- <area>: no issues found

Mark items [x] as they are resolved. An item deliberately left open — not a live bug, a judgment call to defer — still counts as resolved for this purpose: check it [x] and append a **Deferred <date>** — <reason> note instead of leaving it [ ] forever, which would permanently block /status's next-action check with no way to represent "won't-fix" separately from "not yet done." When every item is either fixed or explicitly deferred this way, update the frontmatter to addressed: true.

Notify when done

After saving the report, use the PushNotification tool:

  • Title: Pipeline Review Complete
  • Body: <N> findings (<X> critical, <Y> high). Open docs/pipeline-review/YYYY-MM-DD.md to review.

If there are zero findings, body: Pipeline is healthy — no issues found.

Done when

Report saved to docs/pipeline-review/YYYY-MM-DD.md and push notification sent.

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.