ai-goodies

Collected AI/agentic engineering artifacts from various codebases — ready to adapt for general-purpose software engineering.

What's here

Directory	Contents
templates/	AGENTS.md and CLAUDE.md starters for different project types
skills/	Claude Code slash-command skills (.claude/skills/ or .agents/skills/)
guides/	How-to prose: AGENTS.md, skills, plugins
benchmarks/	A/B harnesses and results for skills, kept out of the skill folders so they aren't installed
examples/	Worked end-to-end sessions showing a skill in use, kept out of the skill folders so they aren't installed

Installing a skill

Every skill is a self-contained folder under skills/ following the Agent Skills open format — plain markdown, no build step, portable to any agent that supports the standard.

Cross-agent, via the skills.sh installer — it detects the agents on your machine (Claude Code, Codex, Cursor, Gemini CLI, 50+ others) and installs into each one's skills directory:

npx skills add omesser/ai-goodies                             # interactive picker
npx skills add omesser/ai-goodies --skill explain-to-manager  # just one skill

Manually — copy the folder into your agent's skills directory. The spec doesn't fix a location, so it depends on the agent: for Claude Code it's ~/.claude/skills/ (user-level) or .claude/skills/ (project-level); .agents/skills/ is a common agent-agnostic convention:

git clone https://github.com/omesser/ai-goodies
cp -r ai-goodies/skills/explain-to-manager ~/.claude/skills/

Then invoke it in a new session — e.g. /explain-to-manager in Claude Code. Skills that need extra setup (e.g. security-review's scanners) document it in their own README.

Skills

Skill	General-purpose?	Notes
security-review/	✅ Yes - in startup/high-tech context	Full scan: Python, Terraform, K8s, JS, Go, Shell, Helm, secrets
release-notes/	✅ Yes	Git + Jira + GitHub PR enriched release notes
reviewer-roulette/	✅ Yes — needs docs/ownership-map.md	Pick one reviewer from git history + ownership map
reindex-docs/	Partial — uses monorepo build scripts	Pattern is general; script invocation is build-tool-specific
linkedin-post-engine/	❌ No — personal content skill (Oded's voice)	Imported from grok/ClawHub; keeps provenance files
expected-value-calculator/	✅ Yes	Finance/investing (not a SWE skill), but fully reusable: self-contained live data via yfinance + deterministic EV math. Imported from grok/ClawHub
explain-to-manager/	✅ Yes	Interview-driven manager briefs: status resets ("it's not ready / AI didn't 100x it"), spend proposals, and defenses of invisible work — trust-preserving, with hard-question prep. Five worked sessions in examples/explain-to-manager/
check-prod-readiness/	Partial — checklist is general; quality gates assume Python + nox	Pre-merge production-readiness gate: config hygiene, code checks, nox tests/lint, pre-commit
scrooge-check/	✅ Yes	Pre-mortem for risky changes, styled as A Christmas Carol: Ghost of Past finds prior incidents via git history, Ghost of Present audits the diff for real risk smells, Ghost of Future writes a satirical incident report tracing back to them — verdict gates go/no-go
daily-news-briefing/	✅ Yes — swap the regions/instruments for your own	User-invoked geopolitics + markets briefing built for scheduled runs: commentary-first search, every item carries its "so what", every market move tagged priced-in or not, trade ideas dropped unless a mispricing is named
developer-voice/	✅ Yes — user-invoked only	Register reset distilled from Google's developer documentation style guide + eng-practices: write like a knowledgeable friend who knows the reader is in a hurry. Substitutions table (simply / please note / sycophancy / ableist idioms), one-idea-per-sentence rule, and per-surface shapes for chat, commits, PRs, code comments, errors, and review feedback. Measured in benchmarks/developer-voice/
bro/	✅ Yes — user-invoked only	Restate the last message in plain language, with no jargon

Note: some skills are imported from the grok/ClawHub marketplace and retain their _meta.json / skill-card.md provenance files.

Templates

File	Best for
AGENTS-general.md	Top-level agent behavior guide for any repo
AGENTS-fullstack-project.md	Projects with multiple convention docs; routing table pattern
AGENTS-minimal.md	One-liner that links to a conventions doc
CLAUDE-context-mode-routing.md	CLAUDE.md for context-mode plugin repos
CLAUDE-python-package.md	Python library: nox, uv, ruff, setuptools_scm
CLAUDE-python-ml-pipeline.md	ML inference pipeline: config, predictors, tests, lint
CLAUDE-fullstack-app.md	CDK + Lambda + PWA; good general workflow preferences section
CLAUDE-fastapi-react-app.md	FastAPI + React + SQLite; RBAC, module system, air-gapped

Guides

Guide	Covers
agents-md-guide.md	Writing a good AGENTS.md / CLAUDE.md; CLAUDE.md vs AGENTS.md
skill-authoring.md	Anatomy of a Claude Code skill, where skills live, and model-invoked vs user-invoked
plugins.md	Installed plugins and how plugin marketplaces work

OSS skills worth installing

These are open-source Claude Code skills — install directly rather than copying.

Skill	Repo	Install	What it does
context-mode	https://github.com/mksglu/context-mode	claude plugin install context-mode --from github:mksglu/context-mode	Routes large-output commands through a sandbox MCP server so raw bytes don't flood context. Includes ctx_execute, ctx_batch_execute, ctx_execute_file, ctx_fetch_and_index, ctx_search tools and a full set of /ctx-* skills.
openhop	https://github.com/naorsabag/openhop	claude plugin install openhop --from github:naorsabag/openhop	Interactive step-by-step data-flow diagram server. Claude pushes YAML flows; you step through them hop by hop. Also in this repo as a copied skill since we had it locally.
ponytail	https://github.com/DietrichGebert/ponytail	claude plugin install ponytail --from github:DietrichGebert/ponytail	Lazy senior dev persona that enforces the YAGNI ladder before building anything.
last30days	https://github.com/mvanhorn/last30days-skill	claude plugin install last30days --from github:mvanhorn/last30days-skill	Researches what people actually said about any topic in the last 30 days across Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, and the web, scored by real engagement (upvotes, likes, market money). Includes a doctor health check for its sources.
drawio-skill	https://github.com/Agents365-ai/drawio-skill	npx skills add Agents365-ai/365-skills -g (or git clone https://github.com/Agents365-ai/drawio-skill.git ~/.claude/skills/drawio-skill)	Turns natural language into editable .drawio diagrams (exports PNG/SVG/PDF/JPG) via the draw.io desktop CLI. 7 presets (ERD, UML, sequence, C4, architecture, ML, flowchart), Mermaid→.drawio, codebase/IaC/SQL→diagram, 10k+ official shapes + AI/LLM brand logos, vision self-check + iterative refinement. Needs the draw.io desktop CLI (brew install --cask drawio).

Skill marketplaces & skill sources

Not individual skills — curated marketplaces, directories, and first-party skill repos worth browsing when you need a new capability.

Source	Who's behind it	What it is	Why it's worth your time
anthropics/skills	Anthropic (official)	Official Agent Skills repo, doubles as a plugin marketplace (/plugin marketplace add anthropics/skills)	The reference implementation: document skills (docx/xlsx/pptx/pdf), skill-creator, and the canonical skill template. Start here.
obra/superpowers	Jesse Vincent (obra)	A full software-development methodology shipped as skills: TDD, planning, debugging, verification	~121k stars; the most thought-per-skill in the ecosystem. Endorsed in depth by Simon Willison. Works across Claude Code, Codex, Cursor, and more.
trailofbits/skills	Trail of Bits	Plugin marketplace of 17+ security skills: CodeQL/Semgrep static analysis, variant analysis, differential code review, audit workflows	The gold standard for security skills, from one of the most respected security research firms.
skills.sh (vercel-labs/skills)	Vercel	The "npm for skills": npx skills add <owner>/<repo> installs into 50+ agents; directory with install-based leaderboard	410k+ installs; the leaderboard surfaces what people actually use, not just what's starred. Best discovery surface.
VoltAgent/awesome-agent-skills	VoltAgent + community	Curated index of 1000+ skills, anchored on official skills from real dev teams (Anthropic, Vercel, Stripe, Cloudflare, Sentry, Hugging Face, Figma…)	The best-maintained awesome-list; filters first-party, thoughtfully-built skills from the sea of bulk-generated ones.

Skill sources - private GitHub accounts

Personally-maintained skill sources — one person's own workflow, published. Higher thought-per-skill than any marketplace; vetted for personal maintenance and activity (as of July 2026).

Source	Who	What it is	Why it's worth your time
mattpocock/skills	Matt Pocock	"Skills for Real Engineers" — straight from his own .claude directory: tdd, grill-me, to-prd, triage…	~158k stars, pushed daily; the category-defining personal skills repo.
addyosmani/agent-skills	Addy Osmani (Chrome eng lead)	24 structured SKILL.md lifecycle skills for Claude Code / Codex / Cursor / Gemini; siblings: web-quality-skills, adverse	~70k stars, active weekly; reference-quality full-lifecycle pack plus a whole portfolio of focused skill repos.
danielmiessler/Fabric + LifeOS	Daniel Miessler	Fabric: the original curated AI prompt/pattern framework. LifeOS (ex-PAI): his personal-AI-infrastructure repo with skills + commands	~43k + ~16k stars, both active this month; the longest-running thoughtful prompt-pattern ecosystem by one person.
wshobson/agents + commands	Seth Hobson	Multi-harness subagent/plugin marketplace (Claude Code, Codex, Cursor, OpenCode)	~38k stars, evolved daily; the canonical personally-maintained subagents collection.
steipete/agent-rules + agent-scripts	Peter Steinberger (founder of PSPDFKit)	Directly copyable rules/commands for Claude Code & Cursor, plus the shared scripts behind his agent setups	~5.7k + ~5.3k stars, pushed continuously; smaller audience, highest thought-per-line.
deanpeters/Product-Manager-Skills	Dean Peters	56 skills + 6 command workflows for product management (discovery, prioritization, PRDs, roadmapping, stakeholder mapping, AI-product strategy); plugin marketplace for Claude Code, Codex, Cursor, n8n	~5.7k stars, pushed weekly; the deepest PM-specific skill library — each skill also teaches the framework's reasoning, not just its output.

Near-misses worth watching: Steve Yegge's beads/gastown agent-workflow stack (~25k, more tool than skill source), Hamel Husain's evals-skills (evals authority, small but sharp), and Armin Ronacher's agent-stuff.

Installed plugins

Plugin	Repo	What it does
context-mode	mksglu/context-mode	Sandbox large-output commands; MCP tools
ponytail	DietrichGebert/ponytail	Lazy senior dev persona; enforces YAGNI ladder
openhop	naorsabag/openhop	Interactive data-flow diagram server

Quick start: new repo setup

1. Copy templates/AGENTS-general.md → AGENTS.md, edit org name and stack.
2. Copy templates/CLAUDE-*.md that fits your stack → CLAUDE.md, strip project-specific sections.
3. Install the skills you want — see Installing a skill.
4. Tune .claude/settings.local.json permissions for the tools that repo actually needs. Run /fewer-permission-prompts to generate an allowlist from your own transcripts rather than copying someone else's.