shipshitdev/skills

typescript-expert

Solves TypeScript/JS problems and refactors TS for type safety: narrowing, branded types, deep-instantiation errors, strict tsconfig, module resolution, JS-to-TS migration.

stack-validator

Validates Biome 2.3+, Bun workspace, Clerk, Next.js 16 and Tailwind v4 config and flags deprecated patterns. Use before stack work or when AI-generated config may be outdated.

domain-modeling

Builds and sharpens a project's domain model. Use when discussing codebase terminology, writing or editing a CONTEXT.md, or recording or editing an ADR.

wayfinder

Charts a foggy effort too big for one session as a map issue of decision tickets, then resolves one ticket per session until the way is clear. Plans only.

codebase-advisor

Survey any codebase as a senior advisor, then hand back either prioritized self-contained implementation plans for OTHER models/agents to execute, or a written architecture and health analysis for humans. Strictly read-only on source code — never implements,…

debug

Debugs failures end to end: builds a repro loop, ranks hypotheses, instruments, and fixes the root cause with a regression test. Use for bugs, crashes, or a fix that failed.

github-inbox

Triages a GitHub work inbox of assigned issues, review requests, mentions, and authored PRs with failing checks. Use when checking what needs attention or prioritizing tasks.

pstack

Playbook orchestrator for verified, unslopped engineering work. Matches a task to a named playbook, applies the principles index, and routes to how, why, architect, arena, swarm, interrogate, tdd, and related skills. Use for pstack, poteto-mode, /pstack, or…

git-cleanup

Audits candidate code and intent against current trunk, then plans or removes only proven-safe or merged branches and worktrees. Defaults to a read-only cleanup plan.

evaluation

Builds evaluation frameworks for agent systems. Use when testing agent performance, validating context engineering choices, or measuring improvements over time.

ask-dev-loop

Ask which Dev Loop skill or flow fits the current situation. A router over the flagship idea-to-ship path.

commit-summary

Writes Conventional Commit messages from git changes, splits unrelated changes, flags breaking changes, and commits after approval. Use when writing messages or committing work.

deslop

Strips AI slop from code, product, and prose: console logs, any types, dead comments, filler copy, dead buttons. Use when asked to unslop, clean AI code, or polish before shipping.

full-code-review

Reviews PR structure, security depth, and devex/flag hygiene with parallel agents after code-review passes correctness. Use for a full or comprehensive PR review.

retro

Reviews a finished agent session for friction and proposes environment fixes (checks, pointers, reviewer rules, steering cuts, skill edits), most severe first. Proposes only.

fix-merge-conflicts

Resolves git merge conflicts correctness-first, then proves the tree still builds. Use when a merge, rebase, cherry-pick, or stash pop leaves conflict markers or unmerged paths.

github-fix-ci

Diagnoses failing or stuck GitHub Actions checks on a PR, finds root cause, and applies targeted fixes, optionally looping until green. Use to fix CI or get a green build.

github-pr-publish

Creates, updates, and publishes GitHub pull requests with clean titles, durable bodies, branch hygiene, and safe push gates. Use when opening a PR or publishing local changes.

pr-comments

Triages a PR's review threads into a severity-tagged digest, or in address mode implements the fixes and drafts per-thread replies for approval. Use for PR review feedback.

performance-expert

Optimizes backend, database, and delivery performance — API latency, queries and indexes, N+1, caching, queues, profiling, bundles, CDN. Not for React render issues.

rules-capture

Detects user preferences and workflow rules in conversation and records them in .agents/memory/captured-rules.md for promotion. Use on always/never or corrections to the AI.

tdd

Drives feature work and bug fixes test-first with red-green-refactor and vertical slices. Use for TDD, test-first implementation, or regression-first fixes.

architect

Sketches types, signatures, and module structure before code, then guides implementation. Use for architect this, or non-trivial work where coding first locks in the wrong shape.

arena

Runs N parallel candidates at the same task, picks a base, and grafts the best parts of the others into it. Use for arena this, or when one attempt would lock in the wrong shape.

blast-radius

Finds what a small change could break elsewhere and proves it with real code. Use for blast radius of X, what could this break, or a small diff you do not trust.

create-verification-skill

Generates a project-local verification skill that drives the app like a user. Use when a repo has no scripted way to prove UI, CLI, or service behavior.

figure-it-out

Designs an auditable playbook when no narrower one fits. Use for a large migration, an ambitious multi-part change, or work a human reviews later. Logs via show-me-your-work.

how

Explains how a subsystem works: architecture, runtime flow, onboarding mental models. Use for how does X work or code walkthroughs. Not for motivation.

interrogate

Runs several independent adversarial reviewers over a diff and synthesizes one verdict without auto-applying fixes. Use to challenge a change or find blind spots.

maintain-verification-skill

Keeps a project's verification skill and feature map accurate using parallel source readers and one live session, ending in at most one PR of proven corrections.

no-comments

Reviews comments on a diff, deletes narration, fixes accepted findings, and offers encodings for claimed constraints. Use before review or when asked to strip comments.

recall

Rebuilds recent working context from chat history, live state, and the shared record into a tight current-state brief. Use to resume or catch up on past work.

setup-pstack

Configures canonical Shipshit Pstack adapters for selected harnesses while preserving existing role choices. Use when setting up Pstack, migrating duplicate installations, or enabling a supported optional adapter.

show-me-your-work

Keeps a TSV decision log (what, why, evidence, result) for long or unattended work. Use for autonomous or multi-phase runs a human reviews afterward.

swarm

Fans out N parallel workers, drains them, and returns one aggregated report. Use for swarm this, parallel coverage, races, gauntlets and exploration partitions.

teach

Explains a change, subsystem or concept in one plain, diagram-built account so the person truly understands it; changes nothing. Use for teach me this or help me understand X.

technical-writing

Applies a layered technical-writing standard (Diataxis, Google style, STE) to docs, RFCs, READMEs, PR descriptions and commit messages. Use when writing or reviewing those.

why

Investigates why code is shaped as it is: design rationale, regressions, postmortems, data-backed thresholds, with cited evidence. Use how for runtime behavior.

interview

Repo-grounded discovery interview that produces a handoff brief for PRD writing, feature intake, or planning.

research

Investigates a question against primary sources and writes one Markdown note citing each claim. Use for docs, API, version or limit facts, or reading legwork to delegate.