agentsclimarketplace

Team audit

Skill simanam/theaiteam/team/skills/meta/team-audit

Sweep theaiteam for staleness, drift, broken links, missing skills, and outdated content. Produce a structured report in _team-gaps.md.From its SKILL.md

Install
npx -y skills add simanam/theaiteam --skill team-audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.8 KB, ~1.5k tokens by cl100k_base, as published. Nobody here has run it

Team audit

A read-only sweep that produces a findings report. Doesn't change anything — that's apply-self-fix's job.

Six audit passes

Run them in order. Stop and surface only after all six are complete.

Pass 1 — Internal consistency

# Find broken markdown links (relative paths only)
cd /Users/amansingh/Documents/theaiteam
grep -rEn '\]\((?!https?://)[^)]+\)' team/ docs/ orchestrator/ adapters/ *.md | \
  while IFS=: read -r file lineno match; do
    link=$(echo "$match" | grep -oE '\]\([^)]+\)' | sed 's/](//;s/)$//' | head -1)
    base=$(dirname "$file")
    target="$base/$link"
    [ -e "$target" ] || [ -e "${target%#*}" ] || echo "BROKEN: $file:$lineno → $link"
  done

Also check:

  • Every role file's default-skills: frontmatter list points at an existing team/skills/<path>/SKILL.md
  • Every workflow's roles: list matches actual files in team/roles/
  • Every command's args line matches what the body describes
  • Every skill's related-skills: resolve

Pass 2 — Frontmatter validity

For every *.md file in team/, orchestrator/, and adapters/*/README.md:

  • YAML frontmatter present (between --- fences)
  • Required fields per type:
    • role: name, description, tools, model
    • skill: name, description, type
    • workflow: name, description, type, triggers, roles
    • command: name, description, type
  • name matches filename (kebab-case)
  • description is non-empty and a real sentence

Pass 3 — Coverage gaps

  • Each role's default-skills list is non-empty
  • Each domain in team/skills/ has at least one skill (currently flagged: devops/, top-level security/)
  • Each brief type value listed in any brief's frontmatter has a matching workflow
  • Each language in team/rules/languages/ has a matching language-reviewer agent in team/agents-extra/language-reviewers/ (or a noted gap)

Pass 4 — Staleness (world drift)

For each version pin found in team/rules/, team/skills/, role files:

  • Language pins: Swift, Kotlin, Python, Node/TypeScript, Go, Rust
  • Framework pins: SwiftUI/iOS target, Compose/Android target, FastAPI, Next.js, React, Tailwind
  • Tool pins: Xcode, AGP, ESLint, Ruff
  • Spec pins: OWASP Top 10 year, GDPR/CCPA references

Procedure:

  1. Grep for the pin (e.g., grep -rEn 'Swift [0-9]\.[0-9]+' team/rules/)
  2. WebSearch / WebFetch the official source to confirm the latest stable version
  3. If pinned version is a major behind: flag as stale
  4. If pinned version is current or one minor behind: OK
  5. Cite the source in the gap entry

Be careful: some pins are deliberate floors ("targeting iOS 16+"). Don't flag floors as stale unless the floor itself is unreasonable (e.g., iOS 12 today).

Pass 5 — Upstream sync drift

Use upstream-sync-check. For each vendored upstream listed in THIRD_PARTY_NOTICES.md, check:

  • Has the upstream had a major release since the vendor date?
  • Have any vendored skills been materially rewritten upstream?
  • Have any new skills been added upstream that we'd want?

Surface as moderate findings — don't auto-pull upstream changes without review.

Pass 6 — Adapter sync drift

# Compare the count of files in team/ vs what was synced into .claude/ etc.
cd /Users/amansingh/Documents/theaiteam
expected_skills=$(find team/skills -name 'SKILL.md' | wc -l)
synced_skills=$(find .claude/skills -type l 2>/dev/null | wc -l)
[ "$expected_skills" = "$synced_skills" ] || echo "DRIFT: $expected_skills SKILL.md files but $synced_skills symlinks in .claude/skills/"

# Check AGENTS.md is fresher than the latest team/ edit
agents_mtime=$(stat -f %m AGENTS.md 2>/dev/null || stat -c %Y AGENTS.md)
latest_team_mtime=$(find team/ -name '*.md' -exec stat -f %m {} \; 2>/dev/null | sort -n | tail -1)
[ "$agents_mtime" -ge "$latest_team_mtime" ] || echo "DRIFT: AGENTS.md is older than team/ — run ./adapters/sync-all.sh"

Output format — _team-gaps.md

Append findings to _team-gaps.md at the repo root. Don't overwrite existing entries; mark resolved ones with status.

## Audit YYYY-MM-DD

### G-NNN — <short title>
**Severity:** trivial | moderate | structural
**Evidence:** path/to/file:line — actual quoted text
**Proposed fix:** what to change
**Citation:** (for staleness) URL of source confirming the drift
**Status:** open | in-progress | fixed-in-<commit-sha> | filed-as-improvement-<brief-slug> | won't-fix
**Logged by:** team-maintainer | <other-role-name>
**Logged date:** YYYY-MM-DD

Numbering: G-001 is the first gap ever logged. Increment globally — don't reset per audit.

When to escalate vs. self-fix

Use apply-self-fix for the rules. Quick guide:

SeverityAction
trivialApply fix immediately. Mark fixed-in-<sha>.
moderateApply if confident; otherwise file an improvement brief.
structuralFile an improvement brief. Run the team-improvement workflow.

Pitfalls

  • Running the audit and "fixing as you find." That mixes audit and apply, which makes commits sprawl. Audit first, fix second.
  • Reasoning about currency from training data without a citation. Always verify.
  • Treating every old version pin as stale. Some are deliberate floors; preserve the intent.
  • Skipping the adapter-sync drift check. It's the most common real-world drift.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,851. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.