Claude Code Slash Commands Pack

24 slash commands for the work you repeat every day in Claude Code, written as modern skills (the format that replaced .claude/commands), each with careful instructions, live git context injected before Claude starts, and only the permissions it needs. Plus a linter that catches the mistakes that make custom commands silently fail, and a cheat sheet of every built-in command.

Names don't clash with any built-in command (checked against the list in the Claude Code docs, Nov 10, 2026). Commands with side effects (/commit, /ship, /handoff) set disable-model-invocation: true, so Claude never runs them on its own. Pushing, opening PRs and running migrations always ask first.

python3 install.py                     # personal (~/.claude/skills): available in every project
python3 install.py --project           # this repo (.claude/skills): commit them to share with your team
python3 install.py --only commit,ship,fix-tests
python3 install.py --uninstall         # removes only the commands it installed

Existing skills with the same name are kept unless you pass --force. Or use it as a plugin (commands become /dev-commands:commit): claude --plugin-dir /path/to/claude-code-slash-commands-pack.

python3 tools/lint_skills.py .claude        # or ~/.claude/skills, a plugin folder, single files
python3 tools/lint_skills.py . --portable   # also check for claude.ai / Skills API uploads

It catches: misspelled or unknown frontmatter fields (Claude Code ignores them without an error), bad booleans/effort/context values, names Claude Code refuses (trailing spaces, invisible characters, <>), names that a built-in already uses, !commands not pre-approved in allowed-tools (the skill aborts outside auto mode), shell pipes in injected commands, ! placed where it won't run, unused declared arguments, skills nothing can invoke, duplicate command names, overly broad Bash grants, and pasted secrets. Exit 1 on errors, so it works in CI.

• CHEATSHEET.md: every built-in command and bundled skill grouped by task, from the Claude Code docs (Nov 10, 2026).
• ./demo.sh lists the commands and lints a sample project with 16 common mistakes. ./run_tests.sh runs 10 tests.

• Commands are instructions for Claude, not scripts: results depend on the model and your repo. They work best with a CLAUDE.md that says how to build and test.
• Injected git context needs a git repository with at least one commit.
• Claude Code changes quickly; the linter's field list and built-in names reflect the docs on Nov 10, 2026 (tools/builtin-commands.txt is easy to update). Not affiliated with Anthropic.

Free companion: Claude Code Statusline Kit: https://youraitools.co/marketplace/claude-code-statusline-kit. Related: Claude Code Safety Hooks: https://youraitools.co/marketplace/claude-code-safety-hooks