Developer-centered AI engineering: build software with security, quality and governance. One local binary under the agent you already run
This repository also configures its own agents. See what ai-engineering tells them →
Skill Claude CodeCodex
Audit the codebase for exploitable security issues — traces attacker-controlled input to dangerous sinks across HTTP handlers, background jobs, auth and access control, and anything shipped to the client, and checks dependencies for known CVEs. Use when asked for a security review or audit, to check for…
Skill Claude CodeCodex
Authors and revises Agent Behavior BEHAVIOR.md specs that capture recurring, judgeable agent conduct. Use when creating, reviewing, splitting, grouping, or calibrating behavior specs, or when translating trace failures and runtime guidance into durable behaviors. Trigger for "write a behavior spec", "turn this failure…
Skill Claude CodeCodex
Test system behavior under extreme load to find breaking points, capacity limits, and failure modes. Covers stress, spike, and breakpoint testing with k6 (primary), JMeter, or Locust. Refuses unbounded concurrency and production saturation without explicit approval. Trigger for "stress test", "load test until it…
Skill Claude CodeCodex
Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports…
Skill Codex
Use when a web interface needs its visual defects found and fixed with measurements rather than opinions — misaligned rows, ragged card interiors, touch targets, rendered contrast over gradients and images, text printing over text, Gestalt proximity, divider lines that should be space, and type-scale drift — measured…
Skill Claude CodeCodex
Audits documentation files against actual repo state. Detects dead references (links to nonexistent files/URLs), false claims (doc says feature X exists but the code doesn't), outdated version numbers, and references to commands that don't exist. Input: path to docs dir or specific files (defaults to docs/ README.md…
Skill Claude CodeCodex
Comprehensive code review skill. Two modes: diff review for over-engineering findings (ranked, compact), and debt collection for deferred shortcut ledgers. Checks against AGENTS.md code style rules (KISS, YAGNI, DRY, early returns, naming). Use when the user asks to "review code", "code review", "find…
Skill Claude CodeCodex
Semantic code intelligence via MCP: call chains, blast radius, cross-file symbol relationships. Uses CodeGraph (colbymchenry/codegraph) — a local-first Rust server that builds a SQLite knowledge graph from tree-sitter ASTs. One tool: codegraphexplore. Use when the question involves "how does X work", "what calls Y"…
Skill Claude CodeCodex
Coordinate a multi-round debate between two agents until the code is clean. adversarial-reviewer (the critic) assumes everything is wrong; fixer implements fixes or disputes them with evidence. They exchange findings through a shared thread file, and each keeps its memory across rounds. Use as the orchestrator's…
Skill Claude Code
Adapts a governed project's architecture rules, config, overrides, and AGENTS.md after ai-eng init. Manual only. Writes a proposal page and stops until the person says yes, then runs ai-eng adapt apply. Trigger for "/ai-config-to-project", "adapt this project", "the init rules are wrong for this repo".
Skill Claude CodeCodex
Interviews the user about their product and visual taste, commits to a specific referenced aesthetic direction, and writes a DESIGN.md design system file in Google's DESIGN.md format — then audits it with the official @google/design.md CLI until it lints clean. Use this skill whenever the user wants a design system…
Skill Claude Code
Build a whole feature end-to-end through subagents. It plans gated checkpoints, writes tests and prototypes, then loops each checkpoint through implement → behavior gate (CLI/code tests, UI e2e last) → UI gate (live vs prototype) → adversarial review gate until every checkpoint passes, then hands it to a human to…
Skill Claude CodeCodex
Build a single-file, clickable HTML prototype of a UI screen or flow, following the design directions given in the request. Use when the user asks to prototype, mock up, or sketch a UI (e.g. "/ai-prototype order list, dense, neutral palette", "mock up the settings page").
Skill Claude CodeCodex
Compare the live UI of the web app against its HTML prototype in .ai-engineering/workflow/prototypes/ and report the visual and behavioral gaps. Screenshots both, then two parallel reviewers (subagents, or claude -p) diff them, one for visuals and one for behavior. Use when the user asks to review, check, or diff the…
Skill Claude CodeCodex
Plan and write the tests for a feature, bottom-up. Everything that code or the CLI can verify (pure logic, API endpoints, DB constraints, auth/role rules) is tested without a browser. UI end-to-end tests come last and cover only what the lower layers can't. Use when the user asks to plan tests, write tests, or add…
Skill Claude CodeCodex
Removes finished local git work in one automatic pass, with no question and no name to type: dead worktrees first (git worktree remove, then git worktree prune), then branches already merged into the local main, then the branches a squash-merged pull request left behind (tree identical to origin/main, tagged before…
Skill Claude CodeCodex
Reads every open issue and pull request of a repository and writes one ranked, actionable page at .ai-engineering/triage.html: what is broken, what is already answered, what to close, and the three things to start on first, each claim backed by a permalink at a commit SHA. Read-only: it never comments, labels, closes…
Skill Claude CodeCodex
The pull-request loop, opened from the local main with no worktree: cut the pull-request branch from the local main, push it, open the pull request against main with a body assembled from the merge commits the branch carries, watch CI and the review comments, fix only what is obvious in bounded iterations, then merge…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: