skills

In Claude Code, run:

/plugin marketplace add officialunofficial/skills
/plugin install officialunofficial-skills@officialunofficial-skills

Or from any shell:

npx skills@latest add officialunofficial/skills

Agent skills for Official Unofficial, Inc. — small, composable slash commands and behaviors for doing real engineering with a coding agent (Claude Code, or any harness that loads skills). Each skill is a directory with a SKILL.md the agent reads to gain a capability; some carry sibling reference files and scripts.

The guiding idea: keep skills small, adaptable, and composable, and keep you in control of the process. They lean on software-engineering fundamentals — deep modules, tight feedback loops, red-green-refactor, a shared domain language — rather than owning your workflow end to end. Fork them, rename them, make them yours.

Layout

skills/
  <category>/
    <skill-name>/
      SKILL.md        # required: YAML header (name, description) + instructions
      *.md            # optional: reference disclosed from SKILL.md, loaded on demand
      scripts/        # optional: deterministic helpers

A skill's name is the slash command that invokes it (/to-invariants). Skills split on one axis — who can invoke them:

• User-invoked (disable-model-invocation: true) — only you, by typing the name. These orchestrate; they cost no context and don't fire on their own.
• Model-invoked — the agent can reach them autonomously when the task fits, and other skills can compose them. These hold the reusable discipline.

New here? Run /setup-skills once per repo, then /which-skill to route to the right tool.

Skills

Engineering

Daily code work. The backbone is a single idea → ship flow — /which-skill maps it.

Flow and orchestration

Skill		What it does
which-skill	user	Router: picks the skill or flow that fits your situation.
setup-skills	user	Configure a repo for the engineering skills: issue tracker, triage labels, domain-doc layout. Run once.
grill-with-docs	user	Relentless interview that sharpens a plan while writing the domain model (CONTEXT.md, ADRs) down as you go.
to-prd	user	Turn the conversation into a PRD and publish it to the issue tracker — no interview.
to-issues	user	Break a plan, spec, or PRD into independently-grabbable issues via vertical slices.
implement	user	Build a piece of work from a PRD or issue, driving /tdd then /code-review.
triage	user	Move incoming issues and external PRs through a state machine of triage roles.

Loops — cycles of work that repeat until a stop condition is met, instead of a single turn.

Skill		What it does
designing-loops	user	Reference for the four loop shapes (turn/goal/time/proactive), writing completion criteria, and managing token usage across a loop's lifetime.
dependency-update-loop	user	Recurring pass over automated dependency-bump PRs: rebase, relock, verify, report what's ready.
stacked-pr-loop	user	Recurring check-in on a stack of dependent PRs: rebase children, surface CI state, flag human-only gates.
deploy-verify-loop	user	Post-deploy loop: poll the rollout to healthy, then run one smoke check against a real user path.

Verification — turn a design or spec into checks that keep an implementation honest. They compound: invariants become fuzz oracles; both sharpen the conformance suite.

Skill		What it does
to-invariants	model	Distil the properties that must always hold: system invariants plus enforceable ones, each mapped to a check or flagged as a GAP.
to-fuzz	model	Distil a fuzzing plan: the targets worth fuzzing, each with an entry point, input model, and failure oracle.
to-conformance	model	Distil a spec's normative requirements into implementation-independent conformance tests, with traceability.

Design, build, and review

Skill		What it does
tdd	model	Test-driven development: a red-green-refactor loop, one vertical slice at a time.
code-review	model	Two-axis review of the diff since a fixed point: Standards and Spec, as parallel passes.
diagnosing-bugs	model	Diagnosis loop for hard bugs and regressions: reproduce → minimize → hypothesize → instrument → fix → regression-test.
codebase-design	model	Vocabulary and principles for designing deep modules: a lot of behavior behind a small interface.
domain-modeling	model	Build and refine a project's domain language; keep CONTEXT.md and ADRs current.
improve-codebase-architecture	user	Scan for deepening opportunities, present them as a visual report, then grill the one you pick.
resolving-merge-conflicts	model	Work through an in-progress git merge or rebase conflict.

Standalone

Skill		What it does
prototype	model	Build a throwaway prototype to answer one design question: a runnable terminal app, or several UI variations.
research	model	Background agent that investigates a question against primary sources and leaves a cited Markdown file.

Productivity

Workflow tools, not code-specific.

Skill		What it does
grill-me	user	Relentless, one-decision-at-a-time interview that sharpens any plan or design.
grilling	model	The reusable interview primitive behind grill-me and grill-with-docs.
handoff	user	Compact the current conversation into a handoff document another agent can pick up.
teach	user	Teach a concept across sessions, using the current directory as a stateful workspace.
technical-writing	model	House style for docs, specs, release notes, and user-facing copy: clear, direct, specific.
writing-great-skills	user	Reference for writing and editing skills well: the vocabulary that makes a skill predictable.

Misc

Skill		What it does
git-guardrails-claude-code	model	Install a Claude Code hook that blocks destructive git commands.
setup-pre-commit	model	Add a Husky pre-commit hook (lint-staged plus Prettier), type checking, and tests.
scaffold-exercises	model	Scaffold an exercise directory tree: sections, problems, solutions, explainers.
migrate-to-shoehorn	model	Convert as assertions in tests to @total-typescript/shoehorn helpers.

mkit

Skills for mkit — the content-addressed, Ed25519-signed VCS. Two for using mkit, three for developing it.

Skill		What it does
mkit	model	Drive the mkit CLI: BLAKE3 object IDs, signed commits, transports — the parts that aren't git.
mkit-attest	model	Produce and verify in-toto/DSSE attestations on a commit against a trust-roots registry.
new-mkit-spec	model	Scaffold a SPEC-*.md in the repo's format, register it in the index, and wire git-parity.
mkit-release	model	Cut a signed, attested release — one v tag drives the GitHub, crates.io, and npm channels.
mkit-ci-preflight	model	Confirm a PR is genuinely green across the GitHub gates and the GCB mkit-*-pr checks before merge.

Installing

In Claude Code, this repo is a plugin marketplace — add it once, then install the whole collection as a single plugin:

/plugin marketplace add officialunofficial/skills
/plugin install officialunofficial-skills@officialunofficial-skills

/plugin update officialunofficial-skills pulls new skills as they land.

Working in another agent, or want to pick individual skills instead of the whole set? Use the skills.sh installer — pick the skills and the agents (Claude Code, Cursor, Copilot, and more) to install them on:

npx skills@latest add officialunofficial/skills

For the full engineering flow, include setup-skills and which-skill, then run /setup-skills once in your agent to configure the issue tracker, triage labels, and doc layout.

Prefer to wire it up by hand? Clone the repo and symlink individual skills:

git clone git@github.com:officialunofficial/skills.git
ln -s "$PWD/skills/skills/engineering/to-invariants" ~/.claude/skills/to-invariants

Then invoke a skill by name (for example /to-invariants, optionally with an output path).

Every SKILL.md here follows the Agent Skills open standard, so the same file works unmodified on other agents that support it — for Codex CLI, symlink into .agents/skills/<name> (project-local) or ~/.agents/skills/<name> (personal) instead of ~/.claude/skills/<name>.

Adding a skill

1. Create the skill directory skills/<category>/<name>/ with a SKILL.md inside.
2. The YAML header needs a name (equal to the directory) and a description. For a model-invoked skill, write the description in third person: the first sentence says what it does, then "Use when …" names the distinct triggers. For a user-invoked skill, set disable-model-invocation: true and keep the description to a one-line human summary.
3. Keep SKILL.md tight; disclose long reference into sibling .md files, loaded on demand.
4. See writing-great-skills for the craft, and technical-writing for the prose style.

Releasing

Releases use CalVer — vYYYY.M.MICRO (for example v2026.7.0, then v2026.7.1, then v2026.8.0). MICRO resets to 0 each calendar month.

To cut one, run the release workflow from the Actions tab. It derives the next version from the date and the existing tags, generates notes from the conventional-commit history, bumps package.json, updates CHANGELOG.md, tags, and publishes a GitHub Release. Run it with the dry-run input first to preview the version and notes.

License

This project is dual-licensed under either of

• the Apache License, Version 2.0
• the MIT license

at your option. Unless you state otherwise, you license any contribution you submit for inclusion under these same terms, without additional conditions.

Credits

Many of these skills are adapted from mattpocock/skills (MIT), with thanks.