claude-skills

Tell the agent to zoom out and give broader context or a higher-level perspective. Use when you're unfamiliar with a section of code or need to understand how it fits into the bigger picture.

Produce a gradeable rubric file from an issue, PRD, or free-form goal — sharpened to the standard required by Anthropic's Managed Agents user.define_outcome event. Sidecar artifact (issues/NNN-*.rubric.md or rubrics/SLUG.md); does not modify source files. Use when the user wants to define "what done looks like" for grader-checkable success criteria. Triggers: /write-a-rubric, "write a rubric", "outcome rubric", "define-outcome", "managed-agent outcome", "make this gradeable".

Interview the user and produce issues/prd.md from a client brief, feature idea, or rough request. Step 1 of the build pipeline (write-a-prd → prd-to-issues → work-issues). Use when no PRD exists yet and one is needed. NOT for breaking an existing PRD into tasks (use prd-to-issues), and NOT for tracker-coupled output (this writes local markdown only). Triggers: /write-a-prd, "write a PRD", "draft a spec", "turn this brief into a PRD".

Autonomously work one task from an issues/ queue end-to-end — explore, implement (TDD when applicable), run feedback loops, commit, update the issue file. Step 3 of the build pipeline (write-a-prd → prd-to-issues → work-issues). Heavyweight (commits code, modifies the issue queue). Use only when explicitly asked: /work-issues, "work the issue queue", "run an AFK iteration", "do one issue end-to-end", OR when another orchestrator skill (e.g. /feature) needs to dispatch an AFK iteration programmatically. Headless loop runner: bin/loop.sh.

Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.

Build a throwaway prototype to flush out a design before committing. Routes between two branches — a runnable terminal app for state/business-logic questions, or several radically different UI variations toggleable from one route. Use when the user wants to prototype, sanity-check a data model or state machine, mock up a UI, explore design options, or says "prototype this", "let me play with it", "try a few designs".

Read an existing PRD (typically issues/prd.md, written by write-a-prd) and break it into vertical-slice tracer-bullet issues at issues/NNN-short-title.md, tagged HITL (needs human) or AFK (autonomous). Step 2 of the build pipeline (write-a-prd → prd-to-issues → work-issues). Use when a PRD exists and needs to become work tickets. NOT for writing the PRD itself (use write-a-prd). Triggers: /prd-to-issues, "break this PRD into issues", "turn this spec into tickets".

Generate or update ARCHITECTURE.md — a living map of a project's structure, dependencies, and high-coupling zones. Polyglot-adaptive. Heavyweight (writes ARCHITECTURE.md, dispatches parallel agents). Use only when explicitly asked: /map, "generate ARCHITECTURE.md", "map this codebase", OR when another orchestrator skill (e.g. /feature step_refresh_map) needs to dispatch it programmatically.

Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Stress-test a plan, design, or idea against the project's domain language (CONTEXT.md) and recorded decisions (docs/adr/). One question at a time, walks the decision tree, captures crystallized terminology and decisions inline. THIS IS THE DEFAULT GRILLING SKILL — use for any interrogation in a real project, even if CONTEXT.md or docs/adr/ don't exist yet (they get created lazily on first term resolution / first ADR-worthy decision). Triggers: /grill-with-docs, "grill me", "grill this", "interrogate this plan", "stress-test this design", "walk me through the decisions", "challenge this". Use grill-me ONLY for pure green-field thinking with no codebase at all — rare.

Pure socratic interrogation with NO project artifacts at all — no codebase, no CONTEXT.md, no ADRs to reference. One question at a time, walk the decision tree, recommend an answer for each branch. Rare edge case — most grilling happens inside a project, so use grill-with-docs by default (it handles the no-CONTEXT.md case by creating one lazily). Use grill-me ONLY when the user is brainstorming entirely outside a codebase (a fresh idea before any code exists, a hypothetical, a non-software design problem).

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when the user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, or describes a performance regression.

Rank functions by CRAP score (complexity × lack of real test coverage) on the current branch, then propose either a refactor or missing tests for the worst offender. Use when the user runs /crap or asks to find risky, complex, poorly-tested code.

Three parallel reviewer agents find contradictions, config gaps, and coverage holes across a project's docs, config, and tests — then fix what you approve. Heavyweight (dispatches parallel agents, can modify files). Supports env-var automation hooks (ADVERSARIAL_REVIEW_TARGETS, ADVERSARIAL_REVIEW_REPORT_ONLY) so an orchestrator can dispatch a programmatic validator pass without keyboard. Use only when explicitly asked: /adversarial-review, "adversarial review", "find gaps in my docs/config", OR when another skill (e.g. /feature) needs to dispatch a validator pass programmatically.