Production-grade Cursor slash commands: each /command ships with a full agent skill contract, a behavioral eval rubric, and CI that fails if safety guards regress. Install as a user plugin and get the same workflows on desktop, web, CLI, and mobile.
Stable release: v1.6.0. Catalog: 44 commands. Not an official Cursor product.
Inspired by install ergonomics from hamzafer/cursor-commands. This repo goes further: paired skills, eval rubrics, and structural ship-gate CI (run-eval-fixtures.py --strict).
- Customize → Plugins → + Add →
https://github.com/emaraschio/cursor-commands(user scope). Reload if/is empty. - In agent chat, run
/define-agent-goalwith a tiny task. Expect a six-part Goal and a halt (no code edits until a later execute now).
Lightly trimmed example (generic names only):
# Agent goal: Reduce flaky tests in service-a
## 1. Outcome
Flake rate under 1% over 10 CI runs on service-a integration specs; no deleted coverage without replacement.
## 2. Verification
### Success criteria
- [ ] Flake rate under 1% across 10 CI runs
- [ ] Existing test intent preserved
- [ ] No production deploys
## 3. Constraints
No production deploys; keep existing test intent.
## 4. Boundaries
service-a/spec/integration/, CI job integration.
## 5. Iteration policy
Max 2 passes: stabilize the top flake, then remeasure.
## 6. Stopping condition
Ask before deleting coverage or changing CI matrix.
Approve or edit this Goal before I execute. Goal approval alone is not permission to execute.
Full install options and duplicate-menu troubleshooting: docs/PLUGIN.md.
Most Cursor command packs are markdown snippets. This one treats prompts like code.
| You get | Why it matters |
|---|---|
| 44 slash commands | Review, TDD, PRs, audits, planning, git sync, and more, ready in / |
| Paired skill contracts | The real workflow lives in SKILL.md, not a one-line prompt |
| Ship-gate evals in CI | Fixtures lock anti-pattern guards so regressions fail the build |
| User plugin package | One install follows your Cursor account (desktop, web, CLI, iOS) |
| Host overlays | Keep employer-specific commands in your workspace; this catalog stays portable |
Portable examples and no org trademarks are a publishing constraint, not the product pitch. The product is reliable agent workflows you can install, fork, and extend.
Full table: COMMANDS_INDEX.md.
| Command | What it does |
|---|---|
/code-review |
Thorough PR review before you approve |
/tdd |
Red-green-refactor with an explicit test-first contract |
/requirement-to-implementation |
Approved requirement → plan → implement → document |
/define-agent-goal |
Turn a rough task into a six-part Goal an agent can run with less babysitting |
/pathfinder |
Fog-of-war planning when the codebase is unfamiliar |
/automation-roi-audit |
Interview real workflows; label Human / AI-assisted / AI-owned; pick a one-week ROI test |
/decision-audit |
Post-build decision ledger and pride gate; no code changes until revise |
/instruction-ablation |
Bare-baseline rebuild of instructions; add only after repeated failures |
/gauntlet-loop |
Beat a real-world quality bar via builder/critic loops |
/design-agent-loop |
Design the loop as a graph: nodes, evidence, retry caps, approval gates, evals |
/merge-open-prs |
Batch review, verify, merge when green (limit 10) |
/thermo-nuclear-code-quality-review |
Extremely strict maintainability pass without changing behavior |
External dependency: /merge-open-prs expects the user-global babysit skill at ~/.cursor/skills-cursor/babysit/SKILL.md. See docs/DEPENDENCIES.md.
- 44 slash commands and 44 skill-contract directories (one pair per command; bodies live under
.cursor/skill-contracts/, not.cursor/skills/) - User plugin package (
.cursor-plugin/+ materializedplugin/tree) for Customize → Plugins - Behavioral evals (
PASS/PARTIAL/FAIL) per command, with ship-gate sections enforced in CI
Recommended: install as a user plugin so commands sync with your Cursor account.
- Open Customize (or Settings → Plugins) and select your user scope.
- Plugins → + Add →
https://github.com/emaraschio/cursor-commands. - Reload the window if
/autocomplete does not list catalog commands yet.
Details, mobile smoke checks, and duplicate-menu troubleshooting: docs/PLUGIN.md.
git clone https://github.com/emaraschio/cursor-commands.git
cd cursor-commands
chmod +x scripts/install.sh
./scripts/install.sh # merge mode (default)Symlinks land in ~/.cursor/commands/ and ~/.cursor/skill-contracts/. Host workspaces resolve contracts from ~/.cursor/skill-contracts/<name>/SKILL.md when the workspace copy is missing. Merge mode keeps your own files; use --replace only to wipe those directories first, or --prune to drop stale symlinks after a rename.
Plugin install and symlink install can coexist but should not: you get duplicate / menu entries. Prefer the user plugin alone, or run ./scripts/install.sh --uninstall first. See docs/PLUGIN.md.
cd path/to/parent-repo
git submodule update --init path/to/cursor-commands
./path/to/cursor-commands/scripts/install.shThe parent repo owns workspace-specific Cursor rules, agents, and memory-bank files. This repo ships shared slash commands and skills only.
CURSOR_HOME="$HOME/.cursor" ./scripts/install.shOrg-specific packs install from the host workspace overlay after the catalog. See docs/ROADMAP.md.
| Doc | What it covers |
|---|---|
| PLUGIN.md | User plugin install, mobile, duplicate / menu |
| EVAL_CI.md | Ship-gate fixtures and CI enforcement |
| VERIFICATION.md | Manual IDE smoke and release records |
| ROADMAP.md | Launch criteria and next priorities |
| COMMAND_SCHEMA.md | Command + skill contract shape |
.cursor/commands/<name>.md # thin slash entry (YAML frontmatter)
.cursor/skill-contracts/<name>/
SKILL.md # full agent contract
eval/
cases.md # behavioral rubric
fixtures.yaml # ship-gate anchors for CI
README.md # how to score
Plugin installs use a materialized plugin/ tree (no .git) from .cursor-plugin/marketplace.json. After catalog edits, run ./scripts/sync-plugin-package.sh.
Schema: COMMAND_SCHEMA.md.
flowchart LR
slash["/command"] --> cmd["commands/name.md"]
cmd --> skill["skill-contracts/name/SKILL.md"]
skill --> eval["eval/cases.md"]
We treat prompts like code: editable contracts need reviewable tests.
- Structural checks in CI: no LLM judge on every PR (flaky and expensive). Ship-gate rows use
fixtures.yaml. - Each command declares ship gate sections in frontmatter (for example
A,S;merge-open-prsusesA,D,E). - Scoring:
PARTIALcounts as fail; target 0FAILon ship gate before merge.
How to run: EVAL_GUIDE.md. Spec: docs/EVAL_CI.md.
python3 scripts/validate-cursor-commands.py
python3 scripts/run-eval-fixtures.py --strict
./scripts/sync-plugin-package.sh # after editing commands/skillsNegative knowledge is captured once, promoted into the skill, and locked in CI:
- Command
## Anti-patternsentry:Trigger / Wrong / Correct / Reason. - Matching positive guard in
SKILL.md. skill_requiredanchor ineval/fixtures.yamlso removing the guard fails--strict.
Entries are curated, not append-only. Optional IDE smoke after install: docs/VERIFICATION.md.
CI on main and PRs: validate, docs, install-smoke, eval-fixtures. Releases: tag v* → release.yml.
Star the repo if this saves you time. Open issues and PRs if you want to extend it.
- Read CONTRIBUTING.md, COMMAND_SCHEMA.md, and docs/EVAL_CI.md.
- Add or change a command as a thin
.cursor/commands/<name>.mdplus.cursor/skill-contracts/<name>/(skill + eval + fixtures). - Run
validate-cursor-commands.pyandrun-eval-fixtures.py --strictlocally before you push. - Walk ship-gate sections when behavior changes; note results in the PR.
Security reports: SECURITY.md. Roadmap: docs/ROADMAP.md.
- Treat prompts like code: skills, evals, and ship-gate CI (dev.to)
- Cursor slash commands docs
- Cursor plugins docs
- hamzafer/cursor-commands: minimal command-only collection
Topics: cursor, cursor-ide, agent-skills, ai-agents, prompts, developer-tools, dotfiles, github-actions, cursor-plugins (see GitHub About on emaraschio/cursor-commands).
MIT. See LICENSE.