Production-grade engineering and output-format skills packaged for OpenAI Codex and Claude Code.
Originally derived from addyosmani/agent-skills, this repo keeps the reusable skill workflows and strips packaging for other agents (Gemini, Copilot, Cursor, Windsurf, opencode, etc.). Both Codex and Claude Code consume the same SKILL.md format. The repo is a marketplace named alanistic-skills that ships two plugins, dev-skills and output-skills, each with both a Codex (.codex-plugin/plugin.json) and a Claude Code (.claude-plugin/plugin.json) manifest.
Clone the repo:
git clone https://github.com/alangmartini/alanistic-skills.git
cd alanistic-skillsInstall one skill on Windows:
$dest = "$env:USERPROFILE\.codex\skills"
New-Item -ItemType Directory -Force $dest
Copy-Item -Recurse .\plugins\dev-skills\skills\test-driven-development $destInstall all skills on Windows:
$dest = "$env:USERPROFILE\.codex\skills"
New-Item -ItemType Directory -Force $dest
Get-ChildItem .\plugins\dev-skills\skills -Directory | Copy-Item -Destination $dest -Recurse -Force
Get-ChildItem .\plugins\output-skills\skills -Directory | Copy-Item -Destination $dest -Recurse -ForceRestart Codex after installing skills.
On macOS/Linux, Codex skills live in ${CODEX_HOME:-$HOME/.codex}/skills:
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
cp -R plugins/dev-skills/skills/* "${CODEX_HOME:-$HOME/.codex}/skills/"
cp -R plugins/output-skills/skills/* "${CODEX_HOME:-$HOME/.codex}/skills/"Claude Code reads SKILL.md files with the same frontmatter Codex uses. Pick one path:
As plugins (recommended). Add this repo as a marketplace and install:
/plugin marketplace add alangmartini/alanistic-skills
/plugin install dev-skills@alanistic-skills
/plugin install output-skills@alanistic-skills
Claude Code auto-discovers skills/*/SKILL.md from each plugin root declared in its .claude-plugin/plugin.json.
As user-level skills. Copy skills into ~/.claude/skills/:
$dest = "$env:USERPROFILE\.claude\skills"
New-Item -ItemType Directory -Force $dest
Get-ChildItem .\plugins\dev-skills\skills -Directory | Copy-Item -Destination $dest -Recurse -Force
Get-ChildItem .\plugins\output-skills\skills -Directory | Copy-Item -Destination $dest -Recurse -Forcemkdir -p "$HOME/.claude/skills"
cp -R plugins/dev-skills/skills/* "$HOME/.claude/skills/"
cp -R plugins/output-skills/skills/* "$HOME/.claude/skills/"As project-level skills. Copy into .claude/skills/ inside your project to share them via version control.
Restart Claude Code after installing or updating skills.
This repo is a marketplace that ships two plugins:
.claude-plugin/
marketplace.json # lists both plugins (marketplace: alanistic-skills)
plugins/
dev-skills/
.claude-plugin/plugin.json # dev-skills
.codex-plugin/plugin.json # dev-skills (Codex)
skills/ # engineering workflow skills
output-skills/
.claude-plugin/plugin.json # output-skills
.codex-plugin/plugin.json # output-skills (Codex)
skills/ # html-output, make-workflow-canvas, make-architecture-flow, make-interactive-visualization, caveman
references/
docs/
The dev-skills plugin covers engineering workflows. The output-skills plugin covers self-contained interactive HTML generation, editable workflow canvases, and the caveman compression mode. Both Codex and Claude Code consume the same SKILL.md format, so each plugin ships both manifests.
Per-agent setup details:
- docs/codex-setup.md — Codex install paths and plugin layout
- docs/claude-setup.md — Claude Code marketplace install, user/project skill folders, troubleshooting
Both Codex and Claude Code read each skill's name and description metadata to decide when to load the full instructions. Keep the description field specific: it should say what the skill does and when the agent should use it.
| Work type | Start with |
|---|---|
| Unsure which workflow applies | using-agent-skills |
| Explicit compressed communication | caveman (output-skills) |
| Handing work to a fresh session | handoff |
| Making the next run of a skill faster or more accurate | improve-skill |
| Starting work on an isolated checkout | create-worktree |
| Vague idea or product concept | idea-refine |
| New project, feature, or significant change | spec-driven-development |
| Turning a spec into tasks | planning-and-task-breakdown |
| Creating an editable workflow builder or node canvas | make-workflow-canvas (output-skills) |
| Creating a codebase architecture map | make-architecture-flow (output-skills) |
| Visualizing diagnostics, incidents, issues, timelines, or evidence packages | make-interactive-visualization (output-skills) |
| Producing a single-file interactive HTML report | html-output (output-skills) |
| Implementing code | incremental-implementation |
| Writing or changing tests | test-driven-development |
| Browser UI verification | browser-testing-with-devtools |
| Broken tests, builds, or behavior | debugging-and-error-recovery |
| Reviewing before merge | code-review-and-quality |
| Security-sensitive work | security-and-hardening |
| Performance-sensitive work | performance-optimization |
| Capturing release notes | create-changelog-fragment |
| Preparing a version release | create-version-release |
| Preparing a release | shipping-and-launch |
| Skill | Purpose |
|---|---|
| using-agent-skills | Discover which skill applies to a task |
| handoff | Compact current context for a fresh agent session |
| improve-skill | Retrospect a finished run for skill and MCP speed and accuracy gaps |
| Skill | Purpose |
|---|---|
| idea-refine | Turn rough ideas into concrete proposals |
| spec-driven-development | Write a structured spec before implementation |
| Skill | Purpose |
|---|---|
| planning-and-task-breakdown | Break specs into small, verifiable tasks |
| Skill | Purpose |
|---|---|
| incremental-implementation | Build in thin, verified slices |
| test-driven-development | Red, green, refactor with evidence |
| context-engineering | Give Codex the right context at the right time |
| source-driven-development | Ground implementation choices in source docs |
| frontend-ui-engineering | Build accessible, production-quality UI |
| api-and-interface-design | Design stable contracts and boundaries |
| Skill | Purpose |
|---|---|
| browser-testing-with-devtools | Verify browser behavior with DevTools MCP |
| debugging-and-error-recovery | Reproduce, localize, fix, and guard failures |
| Skill | Purpose |
|---|---|
| code-review-and-quality | Review across correctness, readability, architecture, security, and performance |
| code-simplification | Reduce complexity while preserving behavior |
| security-and-hardening | Harden auth, input, data, dependencies, and boundaries |
| performance-optimization | Measure first, then optimize bottlenecks |
| Skill | Purpose |
|---|---|
| create-worktree | Create an isolated git worktree and work inside it |
| git-workflow-and-versioning | Keep commits atomic and history useful |
| ci-cd-and-automation | Build quality gates and release automation |
| deprecation-and-migration | Remove or migrate systems deliberately |
| documentation-and-adrs | Document decisions, APIs, and operating context |
| create-changelog-fragment | Write release-ready changelog fragments from completed work |
| create-version-release | Prepare version bumps and release notes from repo history |
| shipping-and-launch | Launch with monitoring, rollout gates, and rollback plans |
| Skill | Purpose |
|---|---|
| implement | End-to-end implementation orchestration (spec, plan, TDD, review, simplify, ship, changelog) |
These ship as a second plugin in the same marketplace, output-skills, living at plugins/output-skills/. Install separately when you want them.
| Skill | Purpose |
|---|---|
| html-output | Produce a single-file interactive HTML report from any source material |
| make-workflow-canvas | Generate a polished editable workflow or architecture canvas with draggable nodes and typed ports |
| make-architecture-flow | Generate an interactive architecture map and agent JSON from a codebase |
| make-interactive-visualization | Generate self-contained interactive visualizations from complex source material |
| caveman | Switch to explicit compressed communication modes |
Claude Code install:
/plugin marketplace add alangmartini/alanistic-skills
/plugin install output-skills@alanistic-skills
Codex install (copy the skills directly):
$dest = "$env:USERPROFILE\.codex\skills"
New-Item -ItemType Directory -Force $dest
Get-ChildItem .\plugins\output-skills\skills -Directory | Copy-Item -Destination $dest -Recurse -Forcemkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
cp -R plugins/output-skills/skills/* "${CODEX_HOME:-$HOME/.codex}/skills/"Reference material is intentionally separate from skill entry points so the agent can load it only when needed:
| Reference | Used for |
|---|---|
| testing-patterns.md | Test structure, naming, mocks, API tests, and E2E examples |
| security-checklist.md | Security review and hardening checks |
| performance-checklist.md | Frontend and backend performance checks |
| accessibility-checklist.md | WCAG-oriented UI review |
New skills belong under the right plugin's skills/ directory: plugins/dev-skills/skills/<skill-name>/SKILL.md for engineering workflows, plugins/output-skills/skills/<skill-name>/SKILL.md for output-format skills. Each skill must include frontmatter with name and description, and the directory name must match the skill name.
Run the validator before committing:
python scripts/validate-skills.pyKeep this repo scoped to Codex and Claude Code. Do not add command folders, manifests, hooks, or setup guides for other agents (Gemini, Copilot, Cursor, Windsurf, opencode, etc.).