officialunofficial-skills

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