Skill Claude CodeCodex
Audit agent-facing documentation (CLAUDE.md, AGENTS.md, copilot-instructions.md) for redundancies, contradictions, gaps, and misplaced content. Supports Claude Code, OpenAI Codex, and GitHub Copilot. For a general agent's system prompt, persona, or tool descriptions (not repo docs), use audit-agent-prompt. Use when…
Skill Claude CodeCodex
Audit general-agent prompts (system prompt, persona, scope, guardrails, tool descriptions, few-shot examples) for framing issues, weak language, missing reasoning, and the operational gaps that cause agent UX failures. LLM-agnostic. Use when the user wants to review or improve a production agent's prompt. For…
Skill Claude CodeCodex
Broad code-quality sweep over a whole codebase. Use when asked to review code health, find technical debt, clean up a codebase, hunt duplication or dead code, or answer what should we refactor — and after a structural change settles, to catch what the refactor left behind. Spins up parallel agents (duplicates…
Skill Claude CodeCodex
Reviews what the comments and docs SAY, not what the code does. Use when asked to clean up comments, docstrings, or README prose; when a codebase feels buried under stale or self-indulgent explanation; when checking whether docstrings still match their signatures; or when hunting machine-written filler. Protects…
Skill Claude CodeCodex
Deep database and SQL audit for relational codebases. Use when asked to review queries, schema, migrations, or indexes; when pages are slow and the database is the suspect; before shipping a migration; or when checking for SQL injection and missing constraints. Thirteen specialized lenses instead of codehealth's…
Skill Claude CodeCodex
Adversarial exploration: roughly 20 agents each trying to break the code from a different angle. Use when asked to stress-test, harden, or attack a codebase; to find edge cases and crash inputs; to answer what breaks this; or after a feature works on the happy path and needs the unhappy ones. Angles include empty…
Skill Claude CodeCodex
Quick sanity check on your current approach. Spawns a fast Sonnet agent to challenge whether you're solving the right problem the right way. Use after building something, before committing to a direction, or when something feels off.
Agent
Scrutinize the codebase at {path} as 4chan /g/ (Technology) would. Style: {style}.
Agent
Scrutinize the codebase at {path} as the Fediverse (Mastodon/Misskey/Pleroma) would. Style: {style}.
Agent
Scrutinize the codebase at {path} as GitHub Issues users would. Style: {style}.
Agent
Scrutinize the codebase at {path} as Hacker News would. Style: {style}.
Agent
Scrutinize the codebase at {path} as LinkedIn tech influencers would. Style: {style}.
Agent
Scrutinize the codebase at {path} as Lobste.rs would. Style: {style}.
Agent
Scrutinize the codebase at {path} as a Reddit community would. Style: {style}. Subreddit: r/{subreddit}.
Agent
This file is executed by the orchestrator (the main Claude Code session), NOT by individual scrutinize agents. The orchestrator reads files once and passes the results to all agents as a snapshot.
Agent
Scrutinize the codebase at {path} as Stack Overflow would. Style: {style}.
Agent
Scrutinize the codebase at {path} as Tech Twitter/X would. Style: {style}.
Agent
Scrutinize the codebase at {path} as YouTube comments on a code review video would. Style: {style}.
Skill Claude CodeCodex
Use when the user wants to stress-test code quality through simulated community reactions (Reddit, HN, Twitter, Lobsters, /g/, Fediverse, Stack Overflow, LinkedIn, GitHub Issues, YouTube) and distill into action points.
Skill Claude CodeCodex
Performance sweep over a whole codebase. Use when asked what's slow, for a performance review, to optimize or speed something up, to hunt latency, memory growth, or find hot spots — and after a feature lands, to catch the cost it added. Spins up parallel agents (hot-loops, allocations, io-batching, blocking, startup…
Skill Claude CodeCodex
Pragmatic DRY review. Finds both under-DRY (true knowledge duplication worth consolidating) and over-DRY (wrong abstractions worth inlining). A single focused agent that makes judgment calls — when to share, when to split, and when to leave it alone.
Skill Claude CodeCodex
Invoke only for the worked bad/good pairs in references/examples.md, or when this session has no earn-the-line rules in context. The rules — prose in comments, docstrings, doc files and prose-valued config fields, plus labels, hints, tooltips, empty states and error text — are normally injected at session start by…
Skill Claude CodeCodex
Reviews what the screen SAYS, not how it looks. Use when asked to review UI copy, labels, hints, tooltips, empty states, error messages, or flash text; when an interface reads as over-explained or self-congratulatory; or when a hint might contradict what the page renders. Covers templates plus the view and domain code…
Skill Claude CodeCodex
Test-suite quality deep-dive: not whether tests exist, but whether they would catch a real bug. Use when asked to review tests, check coverage gaps, judge whether a suite is trustworthy, investigate flaky tests, or work out why a bug shipped past green tests. Spins up parallel agents (coverage-gaps, happy-path-only…