New advertisers: 50% off your first month with code FIRSTMONTHSee slots

Unified skills collection that bundles 16 development-workflow commands into one plugin for Claude Code.

GitHub stars
39
View on GitHub

What it does

  • Adds 16 commands, including /git, /check, /issue, /execute, /qa, /release, and /deploy, each invoked explicitly by the user
  • Covers the full loop: plan an issue, execute it, test, review with /check, open a PR with /git pr, and merge or clean up
  • Enforces explicit invocation by setting disable-model-invocation for Claude Code and implicit invocation off for Codex
  • Designed for developers who want one command per workflow, with action words selecting the playbook, such as /git cleanup

Every skill in skills/ as one plugin

Add marketplace

/plugin marketplace add shipshitdev/library

Run it inside Claude Code once, then install the plugin below.

Install

/plugin install shipshitdev@shipshitdev

Run it inside Claude Code after adding the marketplace.

Files

Discovered from source

Board

One entry point for project boards. The first word of the arguments picks the action; everything after it goes to that action unchanged. Both actions read first and change fields only within an explicit, approved scope.

ActionPlaybookWhat it does
configurereferences/configure.mdAudits a board's fields, columns, and scope, and applies requested setup, copy, or normalization.
syncreferences/sync.mdReconciles a board with current work and delivery evidence and reports gaps; applies only approved field changes.

Dispatch

  1. Read the first word of the arguments. Accept only the two actions above.
  2. Bare /board, or an unknown first word: print the table above and stop. Do not guess an action and do not run any command.
  3. Otherwise read the action's playbook in full and follow it as if it were the skill, passing the remaining arguments as its arguments. Each playbook keeps its own scope, confirmation gates, and final report.

Examples:

/board configure <board URL>        # audit fields, views, and workflows
/board sync <board URL>             # report drift against PR and deploy state
/board sync <board URL> --apply     # prepare approved field corrections

Scope

Loading this router authorizes nothing by itself. Only the selected playbook's contract applies, and its confirmation gates stay in force. Never chain to a second action unless the user asks for it or the selected playbook names that step as part of its own workflow; recommend it instead.

Readme

Ship Shit Dev Skills

Ship Shit Dev Skills

Project Type

16 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.

Quick start

  1. Install the skills (and, optionally, the companion libraries):

    bunx skills add shipshitdev/skills -g --agent claude-code codex cursor --skill '*' -y
    
  2. 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.

How to use

Ship a feature

/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.

Fix a bug

/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 the running app

/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
/qa prod v1.4.0                    # QA a shipped release on production with the QA account, report only
/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.

QA a release, then fix what fails

Claude Code implements; Codex (or any agent) verifies the real app and fixes what breaks.

/qa prod v1.4.0                    # every feature in the release plus the Human Review cards, on production
/verify .tmp/qa-prod/v1.4.0/report.md --surface prod   # debug, fix, and re-verify each failure until green (Codex: $verify ...)
/qa prod v1.4.0 --only F2,F5 --build v1.4.1   # after the fix ships as v1.4.1, re-check the original failures on production
/verify check #231                 # report-only verification of one issue's acceptance criteria

/qa prod signs in as a dedicated QA account, never a real user, with credentials from environment variables and not from the repository. It reports pass or fail per feature and per card with repro steps and evidence, and changes nothing on production beyond records it creates and removes. /verify fixes on a branch, one commit per fix, and never edits production. One-time setup is in skills/qa/references/prod-setup.md.

Review

/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

Land and clean up

/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 and deploy

/release                           # status of trunk, gates, and the next version
/release cut minor
/deploy staging

Daily standup

/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.

Weekly review and retros

/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.

Run several sessions at once

/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

Performance work

/performances parallel             # independent awaits on hot paths -> concurrent, authz and ordered writes stay sequential
/performances dbload               # load-test the schema at 1x/10x/100x/1000x on a throwaway DB, chart before/after
/performances otel                 # OpenTelemetry to the vendor already in use, lint-enforced
/performances logging              # one wide, tail-sampled event per request or job
/performances all                  # all eight, each as its own branch and PR

Each action detects the stack, measures before and after with confidence intervals (scripts/compare.py), keeps behavior unchanged, and opens one PR. Actions that would touch production, like jsonb, only prepare the migration for human approval.

Boards

/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

Commands

CommandActionsWhat it does
/issuefeature (default), prd, plan, check, publish, bugPrepares work as GitHub issues: requirements, plan, readiness gate, publishing, bug reports.
/executeExecutes a prepared issue and tracks review, CI, merge, and deploy evidence.
/testRuns tests at the right scope and reports failures with evidence.
/qarun (default), review, prodInteractive localhost QA loop, a verification pass over finished agent work, or post-release QA on production.
/verifyloop (default), check, debug, fixVerify, debug, fix, and re-verify until green. Runs unattended in Codex or Claude Code.
/checkcode (default), grok, council, weekCode review, a verified Grok second opinion, a scored multi-provider panel, or the weekly engineering review.
/standupall, auditPersonal git recap, or everyone's merged changes with an optional audit.
/gitworktree, safety, conflicts, pr, ci, babysit, modernize, merge, cleanupLocal git and GitHub: worktrees, secret guards, conflicts, PRs, CI fixes, PR babysitting, stale-PR updates, merging, cleanup.
/performancesparallel, dbload, otel, logging, select, url, regex, jsonb, allMeasured, behavior-preserving performance refactors, one PR per action.
/releaseCuts a release from a green trunk.
/deployRuns staging or production deployment workflows.
/boardconfigure, syncAudits and configures GitHub or Jira boards, or reconciles them with delivery evidence.
/orchestratestatus, plan, run, helpCoordinates existing project sessions and routes their questions.
/browseDrives the project's own Brave session over the DevTools protocol.
/imageGenerates raster images through the Codex CLI.
/stackFlags outdated Biome, Bun, Next.js, and Tailwind config.

Renamed commands

BeforeNow
/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

Installation

# 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 -y

Always pass --agent; --all installs into every agent the CLI knows (30+).

Claude plugin

The repository is also a Claude plugin marketplace with one plugin, shipshitdev, that carries every skill in skills/. Add it to your Claude account (Settings → Customize → Skills → Add) to sync the skills to every Claude surface, or in Claude Code:

/plugin marketplace add shipshitdev/skills
/plugin install shipshitdev@shipshitdev

Plugin skills are namespaced, for example /shipshitdev:standup. Codex and Cursor still install with bunx skills add.

Companion libraries

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:

SkillLibraryReach for it when
grillingmattpocockYou want to be interviewed about a plan; /issue prd uses it when installed.
interrogatepstackYou want a multi-model review panel; /check adversarial uses it.
triagemattpocockSorting a pile of incoming issues before filing or fixing.
diagnosing-bugsmattpocockRoot-causing a bug before /issue bug or a fix.
whypstackTracing why code or a decision is the way it is.
retromattpocockLooking back on how a stretch of work went.
reflectpstackTurning a session's lessons into durable rules.
swarmpstackFanning one task out across parallel agents.
arenapstackComparing competing implementations of the same task.

Development

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 tests

To 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.

License

MIT

Details

Marketplace
shipshitdev/library
Category
Community

More from shipshitdev/library

All 140

Similar to shipshitdev

More Development Tools