Cursor Skills Pack: 21 SKILL.md Workflow Skills + Commands-to-Skills Converter

21 ready Agent Skills for Cursor: code review (general, Python, TypeScript/React, Java/Spring), security review, tests, debugging, refactoring, PR descriptions, commit messages, changelog, migrations, dependency upgrades, docs, performance, accessibility, release and staging-deploy workflows. Plus an installer, a converter that turns your old .cursor/commands into skills, and a SKILL.md validator.

Works anywhere SKILL.md folders load: Cursor (.cursor/skills, .agents/skills, .claude/skills, ~/.cursor/skills), Claude Code (.claude/skills) and Codex (.agents/skills). Python 3.9+ for the tools; the skills themselves are Markdown.

./tools/demo.sh                                            # tour
python3 tools/install.py /path/to/repo                     # all 21 -> .cursor/skills/
python3 tools/install.py /path/to/repo --only code-review,security-review,write-tests
python3 tools/install.py --to user                         # ~/.cursor/skills for every project

In Cursor Agent chat type /code-review (or @code-review). Skills with a description that matches your request also load on their own; the four workflow skills that act (pr-description, release-checklist, deploy-staging and any converted commands) only run when you type them.

"auto" = the agent may load it when your request matches its description; "/ only" = disable-model-invocation: true. Path-scoped skills use the paths frontmatter so they only surface for matching files.

Cursor now treats skills as the home for reusable workflows (and /migrate-to-skills exists in Cursor 2.4+). If you'd rather see exactly what changes, or convert commands for another agent, use the converter:

python3 tools/commands_to_skills.py /path/to/repo --dry-run   # preview
python3 tools/commands_to_skills.py /path/to/repo             # writes .cursor/skills/<name>/SKILL.md, keeps the originals
python3 tools/commands_to_skills.py --user                    # ~/.cursor/commands

Converted skills keep command behaviour (disable-model-invocation: true); add --auto to let the agent pick them up on its own.

python3 tools/skill_lint.py .cursor/skills --target cursor checks frontmatter, names vs folders, descriptions, paths, broken file references and more (the same validator is free on its own as the SKILL.md Validator).

./run_tests.sh: every skill passes the validator for the spec, Claude Code, Cursor and Codex with zero warnings; referenced files exist; scripts are executable and run; installer and converter behaviour (skip/force/backup/dry-run, slugging, descriptions, validation). Python 3.13 and 3.10, from a fresh download.

• Skills are instructions: the agent follows them well but not perfectly, and results depend on the model.
• deploy-staging is a template; edit its commands before first use. Nothing here deploys or publishes without you confirming.
• Not affiliated with Anysphere (Cursor), Anthropic or OpenAI.