mattpocock-skills
claudeMatt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).
39 skills
Scale-adaptive agent team for building any project: brief to PRD to architecture to plan to TDD to review to a verified ship. Gates are enforced by hooks and escalate with what the change touches, reports carry the command that reproduces every claim, and work is delegated to subagents so the main context stays small.
Plugin agent-team from marketplace agent-team in LuanNguyen93/Agent-Team. Commands read from the repo's manifest — 2 days ago.
This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.
ai-engineeringBuild and change LLM-backed features where the output is not deterministic - evals instead of a single assertion, model calls treated as expensive non-idempotent I/O, untrusted content entering a prompt as a trust boundary, and cost and latency as budgets. Use when writing or reviewing anything that
app-verifyVerify a change by running the real application and driving it - mobile, desktop, CLI, or HTTP service - for the cases a browser cannot cover. Use before declaring work done on anything that is not a web page.
architecture-disciplineKeeps a change inside the architecture that was agreed - a declared dependency rule, right-sized structure, patterns used only where they earn their place, and algorithms whose cost is stated. Use when designing, planning, writing, or reviewing code.
artifact-templatesTemplates and quality bars for planning artifacts - project brief, PRD, architecture doc, ADR, and user story. Use when producing any of these documents so downstream agents receive a predictable shape.
backend-disciplineServer-side rules that reviews catch late - trust boundaries, transaction scope, idempotency, unbounded queries, and migrations that must survive a rolling deploy. Use when writing or reviewing API handlers, jobs, queries, or schema changes.
brainstorm-grillingSocratic questioning that stress-tests a request before any solution is designed, surfacing the real problem, hidden constraints, and unstated success criteria. Use at the start of a feature or project, before planning.
browser-verifyDrive the running app in a real browser to confirm a change actually works, instead of inferring it from passing unit tests. Use before declaring UI or full-stack work done.
code-navigationFind code by structure rather than by crawling files - query an index or language tooling first, establish the blast radius before changing a shared symbol, and never treat a grep miss as proof of absence. Use before reading, planning, changing, or reviewing unfamiliar code.
context-disciplineKeep the working context small, scoped and truthful - declare what part of the repo this team owns, fold sub-work into subagents, and re-state the constraints that a compaction silently deletes. Use at the start of any session and whenever context grows long.
debug-rcaFour-phase root cause analysis that forbids attempting a fix before the mechanism is understood and reproduced. Use for any bug, test failure, crash, or unexplained behaviour.
design-intelligenceDesign a coherent UI - style direction, colour, typography, spacing, and UX behaviour - so generated interfaces look deliberate rather than machine-assembled. Use when building or reviewing any user-facing screen.
diagram-excalidrawGenerate editable .excalidraw architecture and flow diagrams from the codebase or a design, with a self-review pass for layout problems. Use when a system's shape is easier to show than to describe.
handoff-contractLabel every claim as observed, inferred or assumed, and report in the fixed shape the next agent expects. Use whenever an agent reports back or hands work to another agent.
market-reconReverse-engineers a reference app into a structured recon map (screens S01..Snn, flows F01..Fnn, component inventory, inferred data model, feature matrix CSV) and mines real, linked public user reviews to rank complaints, missing requests, and positioning angles. Use when cloning, competing with, or
obsidian-second-brainDistils verified architectural decisions (ADRs), root-cause analysis (RCA) lessons, domain invariants, and reusable engineering patterns into an Obsidian Second Brain vault as atomic Markdown notes with YAML properties, wikilinks, callouts, and an automated Map of Content (00-MOC.md). Use when captu
pr-createOpen a pull request from the current branch with one gather call and one create call, on GitHub (gh), GitLab (glab) or Azure DevOps (az) as detected from origin. Resolves the base branch, refuses unsafe states (default branch, unpushed or stale branch, suspicious commit range), and writes the descri
pr-descriptionFixed PR description template - What, Why, AC covered, Tests added, Screenshots, Self-review checklist - drafted from the real diff, commits, and this session's gate output.
product-launchPrepares a product or clean-room clone for commercial launch and production cutover - brand naming and trademark screening, landing page and evidence-backed pricing, App Store and Google Play listing linting (listing.py), and the production preflight checklist (preflight.md). Use when naming/rebrand
pulse-reportGenerates evidence-backed Daily and Weekly progress reports across multiple concurrent jobs/projects in one zero-token pass, producing both an Executive Portfolio Overview for you and strictly isolated, copy-paste-ready standup blocks per job.
quality-gatesRun typecheck, lint, test, build, and the project static-analysis or SonarQube gate as a sequential blocking chain where a failure stops the line. Use before marking work complete, before committing, and during review.
react-performanceReact and Next.js correctness and performance rules - re-renders, hooks dependencies, data fetching waterfalls, and bundle cost - ordered by real impact. Use when writing or reviewing React components.
security-disciplineThe security rules that survive contact with production - secrets and rotation, dependency and supply chain risk, authorisation on the object, the browser surface, audit logging and rate limits, and what counts as personal data. Use when planning, writing, or reviewing any code that handles input, i
tdd-disciplineEnforce red-green-refactor when writing or changing code. Requires seeing a test fail for the right reason before any implementation exists. Use whenever implementing a plan, fixing a bug, or adding behaviour.
typescript-disciplineStrict TypeScript discipline and idiomatic type patterns distilled from Matt Pocock's advanced guidelines. Use when designing types, writing robust interfaces, eliminating `any`, or creating reusable generic utilities.
workflow-routerClassify an incoming request into QUICK, FEATURE, or PROJECT and route it through the matching agent-team pipeline. Use at the start of any build request so planning depth matches the actual size of the work.
/agent-team:buildRoute a build request through the agent team at the right planning depth - QUICK, FEATURE, or PROJECT.
/agent-team:prOpen a pull request from the current branch - one gather call, a description in the team's fixed format, one create call on GitHub, GitLab or Azure DevOps.
/agent-team:pulseGenerate a daily or weekly multi-job progress report with an executive portfolio table, isolated copy-paste standup blocks per job, and optional Obsidian Second-Brain archiving.
/agent-team:shipRun quality gates, review on fresh context, verify against the running app, then commit atomically.
/agent-team:specProduce planning artifacts only - brief, PRD, architecture, UI spec - without writing any implementation code.
/agent-team:stack-initDetect this project's stack and gate commands, then write .agent-team.json so the gates run correctly.
/agent-team:upgradeVerify quality gates, sync upstream changes, and self-upgrade Agent Team across Claude Code, Antigravity, and Codex.
analystClarifies a vague, large, or solution-shaped request before any design work begins, and produces a project brief. Use at the start of a FEATURE or PROJECT when the problem is not yet sharply defined. Do NOT use for well-specified small changes or when the user has said to just build it.
architectDesigns the technical shape of a system - components, data model, key flows, failure modes - and records decisions as ADRs with diagrams. Use on a PROJECT after the PRD exists, or when a change needs a structural decision. Do NOT use for changes that fit the existing architecture.
backend-implementerExecutes the backend track of an approved plan using strict red-green-refactor TDD - endpoints, jobs, queries, migrations - against a written contract. Use when a change spans both surfaces and the contract exists, so it can run in parallel with frontend-implementer. Do NOT use for exploration, plan
debuggerFinds the root cause of a bug, test failure, or gate failure through reproduce-locate-explain-fix, and refuses to attempt a fix before the mechanism is understood. Use PROACTIVELY whenever a test fails, a gate blocks, or behaviour is unexplained. Do NOT use for adding new behaviour.
frontend-implementerExecutes the frontend track of an approved plan using strict red-green-refactor TDD - components, client state, screens - against a written contract, stubbing the server until it lands. Use when a change spans both surfaces and the contract exists, so it can run in parallel with backend-implementer.
implementerExecutes an approved plan using strict red-green-refactor TDD, writing the failing test first and keeping the suite green. Use to write or change code once a plan exists, on any change that lives on one side of the client/server boundary. Do NOT use for exploration, planning, or review, and do NOT u
plannerTurns a story or a request into a concrete implementation plan naming real file paths, in dependency order, with a rollback path. Use before any non-trivial code change. Do NOT use for one-line fixes.
pmTurns a project brief into a PRD with numbered functional requirements, measurable acceptance criteria, and a story breakdown. Use after the analyst has produced a brief on a PROJECT. Do NOT use for QUICK fixes or when no brief exists.
qa-verifierRuns the project's quality gates in order and drives the running app to confirm a change actually works. Use PROACTIVELY before reporting any code change with a runnable surface (UI, endpoint, CLI) as done. Do NOT use as a substitute for review, and do NOT use it to fix what it finds.
reviewerReviews a change on fresh context along three axes - compliance with the stated spec, engineering standards, and fit with the agreed architecture. Read-only. Use PROACTIVELY immediately after code is written or modified, always in a context that did not write it. Do NOT use to fix what it finds.
routerRapidly classifies an incoming request into QUICK, FEATURE, or PROJECT tier with a confidence score and suggests the right agent pipeline. Read-only System One decision maker. Do NOT use to plan, route, or dispatch the work it classifies.
ux-designerProduces the design system (tokens, type scale, spacing, colour) and per-screen UI specs including loading, empty, error and edge states. Use when a project or feature has a user interface. Do NOT use for backend-only work or for charts - charts use the dataviz skill.
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/context-budget.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/commit-attribution-guard.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/guard-heavy-load.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/delegation-nudge.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/tier-guard.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/subagent-guard.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/context-budget.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/file-length-guard.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/session-routing.sh
bash "${CLAUDE_PLUGIN_ROOT}"/scripts/gate-task-complete.sh
Matt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).
39 skills
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
15 skills · hooks
Persistent memory for Claude Code: the next session starts where the last one left off.
29 skills · 1 MCP server · hooks
Cross-session memory and token savings for coding agents.
6 skills · 1 agent · 1 MCP server · hooks
Up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your coding agent's context.
1 MCP server
Automates browser interactions for web testing, form filling, screenshots, and data extraction
12 skills
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.