14 user-invoked commands for development workflows. Works with Claude Code, OpenAI Codex, and Cursor.
Every command is one word. Related workflows share a command and take an action
as their first word (/git pr, /issue prd, /check grok); the action's
playbook lives in that skill's references/.
Every skill sets disable-model-invocation: true for Claude Code and
policy.allow_implicit_invocation: false in agents/openai.yaml for Codex.
You run skills explicitly (/git cleanup in Claude Code, $git cleanup in Codex).
Other agents may interpret invocation settings differently. Skills that
compose others read the other playbook directly, so they still work without
exposing those skills to the model.
Skills are model-agnostic playbooks: the harness supplies the model and effort, and
no skill names a concrete model. Standards live in
.agents/memory/system/skill-standards.md.
-
Install the skills (and, optionally, the companion libraries):
bunx skills add shipshitdev/skills -g --agent claude-code codex cursor --skill '*' -y -
Run the core loop on a real task:
/issue add CSV export to the invoices page /execute #123 /check /git pr /git merge
Every command takes free text after its name. /issue, /qa, and /check run
their main action when the first word is not an action; /git, /board, and
/orchestrate help print their actions instead of guessing.
/issue <rough request> # requirements, plan, readiness check, one issue
/execute #<issue> # implement exactly the plan, collect evidence
/test # changed scope; `full`, `e2e`, `types`, `<path>` widen it
/check # Standards + Spec review of the branch or PR
/git pr # push and open or update the PR
/git ci # when checks go red; loops until green on request
Need only one piece? /issue prd interviews you and drafts requirements,
/issue plan turns settled requirements into steps, and /issue check tells you
whether an issue is ready to execute. /issue publish files a finished PRD as an
issue with sub-issues.
/issue bug draft checkout total ignores the discount code # preview only
/issue bug checkout total ignores the discount code # preview, then file
/execute #<bug>
To root-cause before filing, run an upstream debugging skill such as
diagnosing-bugs or /pstack:why.
/qa # branch qa/YYYY-MM-DD off fresh origin trunk, then fix what you report
/qa the onboarding flow # same loop, focused on one area
/qa review # check finished agent work against the ask
/browse open the dashboard and screenshot the empty state
/qa stays in the current checkout and starts a new QA branch from freshly
fetched origin/<default>; with uncommitted changes it stops and lists them
instead. It makes one local commit per fix and never pushes, opens a PR, or
deploys. Publish the branch afterwards with /git pr.
/check # current branch against its base
/check #42 # a PR
/check adversarial #42 # multi-model panel, with pstack interrogate installed
/check grok #42 # second opinion from the Grok CLI, findings verified
/check council #42 # panel of every installed provider CLI, scored rounds in the PR body
/git babysit #42 # watch a PR through CI and review, fix real findings
/git modernize #42 # judge a stale PR, then update it by merging its base
/git merge # review, confirm, merge green PRs, list cleanup candidates
/git merge review # review only, merge nothing
/git merge force # drain the queue: merge green PRs, fix red ones, move on
/git cleanup # dry-run: proves what is merged, deletes nothing
/git cleanup prune # remove the proven-merged branches and worktrees from that plan
/git worktree fix-auth from develop
/git conflicts # resolve the current merge or rebase
/git safety # staged-secret check before a commit
/release # status of trunk, gates, and the next version
/release cut minor
/deploy staging
/standup # your commits, last 24 hours
/standup 7d # your commits, last 7 days
/standup all 24 # everyone's changes merged into trunk, last 24 hours
/standup all 24 audit # same, plus individual and combined review
/standup is report-only. It traces every bullet to a diff and keeps merged
separate from deployed.
/check week # last 7 days, report only
/check week since <SHA> --fix # also repair confirmed findings in scope, via PRs
For a retrospective on how the work went (not what shipped), use the upstream
/retro (mattpocock) or /pstack:reflect.
/orchestrate status # what every project session is doing
/orchestrate plan bugs first # propose lanes, dispatch nothing
/orchestrate run finish the bugs # coordinate existing sessions, routing their questions
/orchestrate help
/board configure <board URL> # audit fields, views, and workflows on a GitHub or Jira board
/board sync <board URL> # reconcile cards with real PR and deploy state
| Command | Actions | What it does |
|---|---|---|
/issue |
feature (default), prd, plan, check, publish, bug |
Prepares work as GitHub issues: requirements, plan, readiness gate, publishing, bug reports. |
/execute |
Executes a prepared issue and tracks review, CI, merge, and deploy evidence. | |
/test |
Runs tests at the right scope and reports failures with evidence. | |
/qa |
run (default), review |
Interactive localhost QA loop, or a verification pass over finished agent work. |
/check |
code (default), grok, council, week |
Code review, a verified Grok second opinion, a scored multi-provider panel, or the weekly engineering review. |
/standup |
all, audit |
Personal git recap, or everyone's merged changes with an optional audit. |
/git |
worktree, safety, conflicts, pr, ci, babysit, modernize, merge, cleanup |
Local git and GitHub: worktrees, secret guards, conflicts, PRs, CI fixes, PR babysitting, stale-PR updates, merging, cleanup. |
/release |
Cuts a release from a green trunk. | |
/deploy |
Runs staging or production deployment workflows. | |
/board |
configure, sync |
Audits and configures GitHub or Jira boards, or reconciles them with delivery evidence. |
/orchestrate |
status, plan, run, help |
Coordinates existing project sessions and routes their questions. |
/browse |
Drives the project's own Brave session over the DevTools protocol. | |
/image |
Generates raster images through the Codex CLI. | |
/stack |
Flags outdated Biome, Bun, Next.js, and Tailwind config. |
| Before | Now |
|---|---|
/file-feature, /write-prd, /write-plan, /check-prd, /publish-prd, /file-bug |
/issue, /issue prd, /issue plan, /issue check, /issue publish, /issue bug |
/run-qa, /review-qa |
/qa, /qa review |
/review-code, /review-with-grok, /review-week |
/check, /check grok, /check week |
/publish-pr, /fix-ci, /merge-prs |
/git pr, /git ci, /git merge |
/configure-board, /sync-board |
/board configure, /board sync |
/execute-plan, /run-tests, /computer-user, /generate-image, /validate-stack |
/execute, /test, /browse, /image, /stack |
/git-cleanup, /release-cleanup, /worktree, /code-review, /weekly-review, /board-sync, /qa-loop |
/git cleanup, /git cleanup, /git worktree, /check, /check week, /board sync, /qa |
# All skills, globally, for the agents you use
bunx skills add shipshitdev/skills -g --agent claude-code codex cursor --skill '*' -y
# One skill
bunx skills add shipshitdev/skills -g --skill git -yAlways pass --agent; --all installs into every agent the CLI knows (30+).
This repository does not vendor other skill libraries. Install them from upstream so you keep receiving their updates:
- pstack (Claude Code plugin from
cursor/plugins):/plugin marketplace add cursor/plugins, then/plugin install pstack@cursor-plugins. - mattpocock/skills:
bunx skills add mattpocock/skills -g --agent claude-code -y.
Skills here name them only as "when installed" and work without them. The ones worth knowing:
| Skill | Library | Reach for it when |
|---|---|---|
grilling |
mattpocock | You want to be interviewed about a plan; /issue prd uses it when installed. |
interrogate |
pstack | You want a multi-model review panel; /check adversarial uses it. |
triage |
mattpocock | Sorting a pile of incoming issues before filing or fixing. |
diagnosing-bugs |
mattpocock | Root-causing a bug before /issue bug or a fix. |
why |
pstack | Tracing why code or a decision is the way it is. |
retro |
mattpocock | Looking back on how a stretch of work went. |
reflect |
pstack | Turning a session's lessons into durable rules. |
swarm |
pstack | Fanning one task out across parallel agents. |
arena |
pstack | Comparing competing implementations of the same task. |
bun install
bun run lint # markdownlint, biome (JSON), shellcheck
bun run validate # scripts/validate.py
bun run test # validator tests and each skill's script testsTo add a command, create skills/<name>/SKILL.md with a one-word name matching
the directory and disable-model-invocation: true. Add agents/openai.yaml with
policy.allow_implicit_invocation: false to keep it explicit-only in Codex too.
See Codex's invocation policy.
Alternatively, add an action to an
existing router as references/<action>.md plus a row in its table. Then run the
three commands above.
Releases are cut by hand with /release cut: a vX.Y.Z tag and a GitHub release.
MIT