Skills

Agent skills that read the receipts: the code, the bill, the ledger, the diff. Trust what you can re-check, cut what you cannot.

Skills

Uncle Bob

unclebob is a do-not-read-the-code workflow for delegated implementation: surround the agent with extreme constraints and read the receipt instead of the diff. The catch that plan usually skips is that the same agent writes the code and the tests. So the skill makes the human approve gherkin scenarios up front, locks failing tests before any implementation exists, forbids mixing code edits and test edits in one round, and runs mutation testing last on both layers, code mutants and mutated gherkin examples, as the test of the tests. Done means a one-table receipt, never a diff to read.

Path: skills/unclebob/SKILL.md

Memory Audit

memory-audit fact-checks an agent's persistent memory against the repo it describes. Every line in CLAUDE.md, AGENTS.md, rules files, and memory directories is a claim that ages: paths move, commands get renamed, flags disappear, and the agent stays confidently wrong. This skill splits the memory into atomic claims, verifies each one against the current codebase and git history, and reports a per-line verdict table (confirmed, stale, wrong, contradiction, preference) with a receipt for every verdict. It reports; the human prunes.

Path: skills/memory-audit/SKILL.md

Engineering Review

engineering-review is a bug-first review stance for diffs, PRs, commits, migrations, architecture changes, implementations, test plans. It focuses on real breakage, missing coverage, operational risk, and user impact instead of style noise.

Path: skills/engineering-review/SKILL.md

Loop Referee

loop-referee referees writer/reviewer loops between coding agents: one tool writes, another reviews, they iterate. It sets a loop contract up front, keeps an append-only per-round ledger, then applies stop rules that end the loop on convergence, disagreement, or divergence instead of burning tokens on polite model deadlock. Built for cross-model setups like Claude Code writing while Codex reviews.

Path: skills/loop-referee/SKILL.md

Skill Auditor

skill-auditor audits a third-party agent skill before you install it. It clones the package read-only, greps for fetch-and-execute, credential access, hidden code, network calls, and broad permission grants, then reports findings with exact file:line receipts and an honest severity. It never installs or runs the thing it audits.

Path: skills/skill-auditor/SKILL.md

Layout

skills/
  engineering-review/
    SKILL.md
  loop-referee/
    SKILL.md
  memory-audit/
    SKILL.md
  skill-auditor/
    SKILL.md
  unclebob/
    SKILL.md

Flat by design. No category maze.

Install

Any agent (Claude Code, Codex, Cursor, Copilot, Gemini, Windsurf, more):

npx skills add brenbuilds1/skills

Claude Code, as a plugin:

/plugin marketplace add brenbuilds1/skills
/plugin install bren-skills@bren-skills

Manual: copy or symlink any folder under skills/ into your agent skills directory.

cp -R skills/unclebob ~/.codex/skills/