craftsmanship

September 6, 2026 · View on GitHub

CI License: MIT

Skills that encode engineering discipline -- think before coding, verify before claiming done, ship with care.

These are agent-agnostic skills compatible with 75+ AI coding agents (Claude Code, Cursor, Codex, Cline, Gemini CLI, GitHub Copilot, and more) via the skills ecosystem.

Install

npx skills add sagargupta16/craftsmanship

The CLI detects which coding agents you have installed; if it finds none, it prompts you to pick. The run is interactive by default: you confirm what gets installed, and you choose between symlinking into each agent's skills directory (upstream's recommended method) and copying real files. -y skips the confirmations, --all installs every skill to every agent without them, and --copy forces real files. When an agent runs the command rather than a human, the CLI detects that and installs non-interactively.

Other flags worth knowing:

npx skills add sagargupta16/craftsmanship -g   # user-level instead of project-level
npx skills update                              # refresh installed skills
npx skills list                                # see what is installed

The Skills

SkillWhen It Activates
planImplementing features spanning 3+ files, unclear requirements, or multi-step work
guardGit operations, handling secrets, or running destructive commands
debugA test fails, a bug is reported, or a fix did not work
verifyBefore claiming work is done, before opening a PR
reviewReviewing code changes, examining PRs, checking your own work
auditSetting up a new repo or auditing an existing one for hygiene issues
shipPreparing to ship -- opening a PR, cutting a release, deploying

What a Skill Looks Like

Each file is YAML frontmatter plus a markdown body written to be read by an agent mid-task. The tables are the load-bearing part -- they turn a vague instruction ("verify it works") into a lookup. skills/verify/SKILL.md opens with a Change Type to Minimum Verification table, so an agent that just finished a database migration reads "Run migration, verify schema, test rollback" instead of guessing.

The description field in the frontmatter is what the agent matches against, so it names the triggers explicitly ("Use when completing a task, before claiming work is done, or before creating a PR"). The body is what the agent reads once triggered.

Philosophy

Four rules run through the skills:

  1. Think first for big changes. A plan for a 10-file feature costs 15 minutes; debugging the wrong implementation costs a day.
  2. Never claim done from tests alone. Tests verify code correctness, not feature correctness. Actually run the thing.
  3. Reproduce before you theorize. A bug you cannot trigger on demand is a bug you cannot confirm you fixed.
  4. Small reversible actions are free. Destructive actions need confirmation. rm -rf, git push --force, DROP TABLE -- pause and confirm.

Install Individual Skills

Use --skill to install a subset instead of the whole set:

npx skills add sagargupta16/craftsmanship --skill plan
npx skills add sagargupta16/craftsmanship --skill debug --skill verify --skill review
npx skills add sagargupta16/craftsmanship --list   # see what is available first

--agent narrows the install the same way (--agent claude-code --agent cursor), and --agent '*' targets every agent the CLI knows about.

Compatible Agents

Installed skills work with any agent that reads .agents/skills/ or an agent-specific directory. The CLI handles the fan-out:

  • Claude Code (.claude/skills/), Cursor, Codex, GitHub Copilot, Gemini CLI, OpenCode
  • Cline, Windsurf, Zed, Warp, Amp, Antigravity, Continue, Kilo Code, Roo Code, Trae
  • Augment, Goose, Droid, Junie, Qwen Code, Grok Build, Kiro CLI, OpenHands
  • Deep Agents, Firebender, Kimi Code CLI, Devin for Terminal
  • Any other agent that supports the SKILL.md format

The full, authoritative list of agents and the exact install paths each one uses lives in the upstream supported agents table.

Structure

craftsmanship/
├── README.md
├── CONTRIBUTING.md
├── CHANGELOG.md
├── CLAUDE.md
├── LICENSE
├── .gitignore
├── renovate.json
├── .github/workflows/ci.yml
└── skills/
    ├── plan/SKILL.md
    ├── guard/SKILL.md
    ├── debug/SKILL.md
    ├── verify/SKILL.md
    ├── review/SKILL.md
    ├── audit/SKILL.md
    └── ship/SKILL.md

Each skill is self-contained -- a single SKILL.md with YAML frontmatter and markdown body.

Contributing

Issues and PRs welcome -- a new skill, a gap in an existing one, or a wrong claim to fix.

See CONTRIBUTING.md for the proposal flow, the required frontmatter, the house format, and how to lint a skill before opening a PR.

Also By This Author

License

MIT. See LICENSE.