skills

Referee for writer/reviewer loops between coding agents, where one tool writes code and the other reviews it in rounds: Claude Code writing while Codex reviews, or the reverse. Use when running or setting up a cross-model review loop, a test-fix loop, or any "loop until clean" workflow. Sets the loop contract up front, keeps an append-only per-round ledger, applies stop rules so the loop converges, ends, or escalates to a human instead of burning tokens on two agents disagreeing. Does not write or fix code itself.

Audit a third-party agent skill (Claude Code / Codex / Gemini CLI SKILL.md packages) before you install it. Use when asked to audit, vet, review, or security-check a skill, plugin, or marketplace package, or before installing one from GitHub or a marketplace. Clones the package read-only, runs supply-chain greps for fetch-and-execute, credential access, hidden code, network calls, and broad permissions, then reports findings with exact file:line and an honest severity for each. Does not install anything.

Personality test for a notes collection. Use when asked to type a vault, analyze a second brain, run a personality test on notes, or answer "what do my notes say about me". Works on an Obsidian vault or any folder of markdown notes. Measures countable behavior only (reopen rate, orphan rate, dated filenames, burst gaps) and assigns one of eight types with the numbers as receipts. Types the notes, not the person. Read-only; never deletes or rewrites anything.

Code review stance for serious engineering changes. Use when asked to review a diff, PR, commit, branch, patch, architecture change, migration, implementation, or test plan for bugs, regressions, missing tests, operational risk, security/privacy issues, and maintainability problems that could actually hurt users.