agent-skills

Explore a codebase to surface architectural friction and propose refactors toward deep modules (simple interface, large implementation) as GitHub issue RFCs. Use when the user asks to audit architecture, find structural friction, or identify refactor opportunities across a codebase. Don't use for single-module interface design (use /codebase-design) or code-level review of a diff (use /code-review).

Write end-to-end tests for user flows using Cypress. Use when user asks to "write e2e tests", "/e2e", "add Cypress tests", or wants to test a user flow end-to-end. Don't use for unit tests, component tests, or projects using Playwright, Puppeteer, or other non-Cypress frameworks.

Audit internationalization coverage and find hardcoded strings. Use when user asks to "check i18n", "/i18n", "find hardcoded strings", "check translations", or wants to verify translation coverage. Don't use for backend string extraction, non-frontend code, or projects without an i18n library.

Check for sensitive data in repository. Use when user asks to "check for sensitive data", "/safe-repo", or wants to verify no company/credential data is in the repository. Use `--diff` mode to scope to staged + unstaged changes only (e.g., before commit). Don't use for general code review, adding .gitignore entries, or scanning non-git directories.

Implement one phase of a plan — reads the plan, finds the next incomplete phase, implements it with feedback loops, marks checkboxes, offers a commit (one phase per invocation). Use after /plan when a plan exists at `.specs/plans/<slug>.md`, or when the user asks to build or implement the next phase. Don't use for ad-hoc changes with no plan (use /tdd) or to check progress (use /test).

Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.

Record an Architecture Decision Record (ADR) — a 1–3 sentence note capturing what was decided and why. Use when user says "create an ADR", "record this decision", "/create-adr", or just decided something architecturally significant. Don't use for forward-looking specs (use /spec) or general repo conventions (use CLAUDE.md).

Harden code proactively against vulnerabilities at the boundary where untrusted input enters the system. Use when implementing auth, handling user input, storing or transmitting sensitive data, integrating external APIs, adding file uploads, or any code that crosses a trust boundary. Don't use for reactive secret scanning (use `safe-repo`) or dependency CVE checks (use `deps-audit`).

Commit and push changes with a pre-launch gate (validate-code + safe-repo) by default. `--fast` skips the gate.

Implement features using official docs for exact dependency versions instead of training data. Use when user asks to "check the docs", "use official docs", "source-driven", or when implementing with unfamiliar APIs/libraries. Don't use for well-known stdlib APIs or project-internal code.

Validate UI changes in a real browser using Chrome DevTools or Playwright MCP. Takes screenshots, compares before/after, exercises interactions, captures console errors. Use when user asks to "visual validate", "/visual-validate", "check the UI", "screenshot before/after", or finishes a UI change. Don't use for unit tests (use `tdd`), E2E user flows (use `e2e`), or backend changes.

VERIFY phase — prove the code works. Run validation (lint, types, tests) plus coverage on changes, and when a plan exists verify its checkboxes against the codebase. Use after /build, when the user asks to verify, test, or confirm work holds up, or to check plan progress. Don't use to implement phases (use /build) or write new tests from scratch (use /tdd).

Actively build and sharpen a project's domain model — challenge terms, pin down a ubiquitous language, and write the glossary and decisions down as they crystallize. Use when the user wants to define domain terminology, resolve ambiguous terms, or when another skill needs the shared vocabulary. Don't use for recording a single architectural decision (use /create-adr) or explaining existing code (use /explain-code).

Investigate a question against primary sources and capture the findings as a cited Markdown file. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background agent. Don't use for implementing against pinned dependency versions (use source-driven) or searching the current codebase (use grep/explore).

Resolve an in-progress git merge or rebase conflict by recovering each side's intent, then finishing the merge. Use when a merge or rebase is mid-conflict, git reports conflicted files, or the user asks to resolve conflicts. Don't use for preventing future conflicts, general git operations, or rewriting history where nothing is conflicted.

Compact the current conversation into a handoff document for another agent to pick up. Use when the user wants to hand off, transfer context to a fresh session, or says "write a handoff". Don't use to summarize for the current session (just answer) or to brief on plan progress (use /test).

A relentless interview to sharpen a plan or design. Use when the user wants to stress-test a plan, pressure-test assumptions, or says "grill me" or "poke holes in this". Don't use for code review (use /code-review) or gathering initial requirements (use /spec).

Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, or build a new skill, or asks "make a skill for X". Don't use for benchmarking or optimizing a skill's triggering with evals (use skill-creator), or for one-off prose edits to an existing skill.

Check dependencies for known vulnerabilities and staleness. Use when user asks to "audit dependencies", "/deps-audit", "check for vulnerabilities", or wants to check dependency health. Don't use for reviewing code quality.

Run linting and formatting checks. Use when user asks to "run linter", "/lint", "check linting", "fix lint errors", or requests code linting/formatting. Don't use for running tests or type-checking only.

Validate code quality: auto-fix formatting/lint, verify types, run tests. Use when user asks to "validate code", "/validate-code", "check code", or wants to validate before committing. Don't use for committing, pushing, or writing new tests.

Group unstaged changes into atomic commits by concern, following repository style. Use when user asks to "commit", "create a commit", "commit changes", or "/commit". Don't use for pushing (that belongs to /ship) or creating pull requests.

Simplify working code without changing behavior — reduce complexity, remove dead abstractions, favor clarity over cleverness. Use when code works but is harder to read or maintain than it should be, or the user asks to "simplify", "clean up", "reduce complexity", or "make this clearer". Don't use to add features or backfill tests (use /tdd or /fortify), design a new interface (use /codebase-design), or review a PR (use /code-review).

Turn a spec into a multi-phase implementation plan using tracer-bullet vertical slices. Use after /spec when a spec exists at `.specs/specs/<slug>.md`, or when the user asks to break work into phases or slices. Don't use without a spec, or for single-file changes with obvious scope.

Orchestrated REVIEW phase — fan out parallel, read-only reviewers over a diff (scope-detected audit skills AND independent agent lenses), then consolidate into one severity-ranked verdict. Use for a full pre-ship review, "review/validate this PR", or "spawn agents to review". Don't use to implement fixes (use /build or /tdd), for a single focused audit or one-lens pass (call it directly, e.g. /a11y-audit, /code-review), or to triage existing reviewer comments (use /triage-review).

Create a spec (PRD) through user interview, codebase exploration, and module design. Use when starting a feature with unclear requirements, when the user asks to spec or define what to build, or says "write a spec" or "write a PRD". Don't use when requirements are crisp and a plan already exists (use /plan or /build).

Front-load UI construction decisions — component boundary and prop API, where state lives, data/async, the required UI states (loading/empty/error), responsive behavior, and accessibility by construction. Use when creating or refactoring a component, page, or design-system primitive. Don't use for pure logic (tdd), throwaway direction exploration (prototype), generic module design (codebase-design), or auditing finished UI (a11y-audit/perf-audit/visual-validate).

Fetch unresolved review comments on a GitHub PR (Copilot bot + human reviewers), verify each against the current code, and classify as Address / Skip / Optional / Discuss with a one-line rationale. Use when the user asks which review comments to address, to triage Copilot/reviewer feedback, sort signal from noise on a PR, or decide what makes sense to fix and what not. Don't use to reply to or resolve threads (out of scope — this is read-only), or to generate a fresh review (use /code-review or /review).

Explains code with visual diagrams and analogies. Use when explaining how code works, walking through a codebase area, or when the user asks "how does this work?" Don't use for modifying code, fixing bugs, generating new implementations, or multi-session concept learning (use /teach).

Build a throwaway prototype to flesh out a design — 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 explore or compare design directions before committing, or says "prototype", "mock up", or "try a few approaches". Don't use for production implementation (use /build or /tdd).

Structurally edit and improve article drafts — reorder sections, tighten arguments, improve clarity. Use when user asks to "revise", "improve my article", "edit my draft", or "/revise". Don't use for typo fixes or formatting (use prose-fix), code documentation, or non-article content.

Teach the user a new skill or concept within this workspace — stateful lessons, references, and learning records tied to a mission. Use when the user asks to learn, be taught, or understand a concept or technology in depth. Don't use to explain a specific code area (use /explain-code) or answer a one-off question.

Audit frontend bundle size and performance. Use when user asks to "audit performance", "/perf-audit", "analyze bundle", "check bundle size", or wants to find performance bottlenecks. Don't use for backend performance, database query optimization, or projects without a frontend build step.

Review a GitHub Pull Request for bugs, security, performance, and code quality. Use when user asks to review a PR or wants pull request feedback. Don't use for reviewing local uncommitted changes, creating new PRs, or merging branches.

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

Fortify existing code by splitting large functions, adding edge-case coverage, and backfilling unit tests. Use when user asks to "fortify", "harden", "bulletproof", "make robust", "make solid", "strengthen", "add missing tests", "split functions", or wants to improve reliability of existing code. Don't use for new features (use tdd), refactoring plans, or code review (use code-review).

Fix prose formatting, typos, and clarity issues in markdown or text files. Use when user asks to "fix dashes", "fix typos", "clean up text", "improve sentences", or "/prose-fix". Don't use for code style, linting, or full rewrites.

Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.

Create a GitHub pull request — fills the repo's PR template, conventional-commit title, ticket refs. Draft mode via `--draft`.

Set up Husky pre-commit hooks with lint-staged (Prettier), type checking, and tests. Use when user asks to "add pre-commit hooks", "setup husky", "setup pre-commit", "configure lint-staged", or wants commit-time formatting/typechecking/testing. Don't use for running linters manually or writing tests.