claude-skill-orchestration-audit
Skill15 starsChanged 9 days agoMITView on GitHub
Audit skills for agentic orchestration quality. Checks subagent dispatch patterns, parallelism opportunities, anti-patterns, and alignment with the three-phase pattern (setup → dispatch → aggregate). Use when reviewing skill quality, saying "audit orchestration", "check my skills for parallelism", "skill orchestration review", or "/claude-skill-orchestration-audit".
In this repository31 other files it gives its agents.
- claude-permissions-auditSkill
- claude-skill-prereq-auditSkill
- claude-skill-spec-auditSkill
- dev-cli-consistency-auditSkill
- dev-code-qualitySkill
- dev-new-toolSkill
- dev-software-designSkill
- gh-dep-pr-triageSkill
- gh-issue-reportSkill
- gh-org-chartSkill
- gh-oss-go-bumpSkill
- gh-oss-release-prepSkill
- gh-oss-releaseSkill
- gh-review-triageSkill
- git-conventional-commitSkill
- gws-meeting-schedulerSkill
- p-activity-digestSkill
- p-blog-post-miningSkill
- p-blog-writerSkill
- p-clipboard-copySkill
- p-daily-reflectionSkill
- p-daily-reportSkill
- p-daily-standupSkill
- p-ego-searchSkill
- p-md-to-slidesSkill
- p-meeting-prepSkill
- p-news-briefingSkill
- project-namerSkill
- p-slack-triageSkill
- p-violentmonkey-userscript-syncSkill
- restaurant-searchSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Check skills against orchestration best practices in references/subagent-orchestration-best-practices.md.
- Very specific: line 32
Read references/subagent-orchestration-best-practices.md — defines the rules
- Very specific: line 33
Read references/agent-skill-analyzer.md — defines the per-skill analysis task
- Distinctive: line 47
Look for phases that iterate over N independent items — even if written as sequential steps:
- Distinctive: line 73
After all agents return:
- Distinctive: line 114
After fixing, re-run the check and show updated table.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill Orchestration Audit
Check skills against orchestration best practices in references/subagent-orchestration-best-practices.md.
Arguments
- Skill names: filter to specific skills
- --fix: suggest and apply corrections interactively
- Default: scan all skills authored by the user (
metadata.authormatch)
Step 1: Setup
- Read
references/subagent-orchestration-best-practices.md— defines the rules - Read
references/agent-skill-analyzer.md— defines the per-skill analysis task - Glob
~/.claude/skills/*/SKILL.mdto discover skills. If a directory path is provided as argument, glob<path>/*/SKILL.mdinstead - Filter to skills where
metadata.authormatches the user (default:jackchuka). If skill names provided, filter to those instead - Classify each skill into one of these categories:
| Category | Audit type | Signal |
| Multi-source aggregation | Orchestration quality | Already dispatches 2+ parallel agents against independent sources |
| Multi-step workflow | Latent parallelism scan | Sequential phases where some sub-tasks within a phase are independent |
| Single-source operation | Skip | One tool/API, truly linear flow, no fan-out possible |
| Meta/utility | Skip | Routers, reference libraries, formatters |
How to detect latent parallelism in multi-step workflows:
Look for phases that iterate over N independent items — even if written as sequential steps:
| Pattern | Example | Signal |
| Per-item scaffolding | "scaffold X for each actor/flow/feature" | N independent CLI calls |
| Per-item implementation | "write one file per resolver/page/command" | N independent file writes to separate paths |
| Multi-source research | "search Odoo, Oracle, SAP" | N independent web fetches |
| Independent tracks | "implement commands" then "implement queries" | Parallel tracks after a shared dependency |
A skill is "multi-step workflow" (audit it) if any phase has 2+ items that don't depend on each other. If every phase is truly linear with no fan-out, classify as "single-source operation" and skip.
- Set
SKILLS_TO_AUDIT= all skills classified as "multi-source aggregation" or "multi-step workflow"
Step 2: Dispatch Agents (parallelize if possible)
Launch one agent per skill in SKILLS_TO_AUDIT. Each agent receives:
- The skill's SKILL.md content
- The skill's
references/directory listing (if any) - Its classification from Step 1
- The check rules from
references/agent-skill-analyzer.md
Each agent returns structured findings per the format in the reference file. If a skill's SKILL.md can't be read, skip gracefully.
Step 3: Aggregate & Report
After all agents return:
- Merge findings across all skills
- Sort by severity: ERROR → WARN → INFO
- Produce single consolidated report:
## Orchestration Audit
### Skills with existing orchestration
| Skill | Category | S1-4 | P1-4 | C1-3 | A1-3 | X1-4 | Issues |
|-------|----------|------|------|------|------|------|--------|
### Skills with latent parallelism opportunities
| Skill | Opportunity | Phase | Benefit | Issues |
|-------|-------------|-------|---------|--------|
### Findings
#### [skill-name]: [issue count] issues
- **[RULE]:[SEVERITY]** — [description]
- Current: [what the skill does now]
- Suggested: [what it should do]
### Top N Highest-Impact Improvements
Rank by benefit (high > medium > low) and usage frequency.
### Summary
N skills checked, M have orchestration, P have latent parallelism, X issues found (E errors, W warnings, I info)Fix (if --fix or user asks)
| Severity | Action |
| ERROR | Show the fix, apply with Edit after user confirms |
| WARN | Show suggestion, apply only if user agrees |
| INFO | Note for awareness, no auto-fix |
After fixing, re-run the check and show updated table.
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
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.

