mattpocock-skills
claudeMatt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).
39 skills
Disciplined plan -> spec -> ship workflow for Claude Code: adversarial review against pinned commits, briefs-not-transcripts context hygiene, and a full paper trail on a pluggable issue tracker (Jira or GitHub Projects).
Plugin sy from marketplace shipyard in nearmap/shipyard. Commands read from the repo's manifest — 7 days ago.
This plugin runs code on your machine — its hooks fire automatically on events like tool use and it starts local MCP servers. The exact commands are listed below; read them before installing.
ciTriage the current branch's CI. Find the relevant current-head run, wait if pending, diagnose failures without flooding context, compare against main when appropriate, and report or fix. In fix mode, push and re-watch until the new current HEAD is green. Never merge.
configInspect, validate, change, and reload this repo's Shipyard configuration. Shows every resolved value with the layer it came from, explains why a value is what it is, writes a setting at the right scope, and re-resolves mid-session so an edit takes effect on the next dispatch without a restart. Never
explainTeach one gnarly topic (a bug family, a design constraint, a system) until you genuinely understand it: author a verified, layered explainer doc through an isolated investigation, then walk it with you one layer at a time, with checkpoints, live repros on challenge, and a decision — never a lecture
helpAnswer questions about Shipyard itself — commands, SY_*/ACLI_* config knobs, which model an agent uses, workflow behavior, diagnosis recipes — by reading the plugin's own docs and citing the source. Never reads the tracker or the target repo's code.
init-repoGet this repo's Shipyard config from zero (or partially done) to genuinely usable: write `.shipyard/config.json` (shared, tracked) and `.shipyard/config.local.json` (personal, gitignored), retire any leftover `SY_*` variables from a legacy `env` block, then prove it live with the same preflight chec
mvpBuild a working MVP fast on a branch that never merges, outside spec, GATE, and merge standards, with every shortcut recorded as it is taken; then harvest it into PR-sized Tasks that each port by hand from a pinned commit on that branch. Unlike /sy:spike, which measures an idea and returns a verdict
planTurn a large objective or existing Epic into a living roadmap with a configured cap (plan.max_active_tasks) of active /sy:spec-ready Tasks, real seams, dependency order, and flat tracker execution.
prCreate, promote, or clean up the GitHub PR for the current branch; keep the description brutally short, preserve durable acceptance evidence in comments, and handle review threads. On an explicit `merge`, merge it once CI is green and reviewers are settled.
shipExecute one fully planned Task to a reviewable PR with isolated writers, immutable independent review, fresh CI/review coverage, and durable accounting.
shutdownWind down one ticket's footprint on the user's request: stop lingering background tasks, remove its recoverable worktrees and branches, close the ticket. Anything unrecoverable or ambiguous is kept and asked about. For an MVP ticket, also release its environment hold and close it as decomposed. Use
specTurn a fuzzy goal or existing Task into one self-contained, execution-ready Task plan. Scope the surface, resolve standards, research load-bearing behaviour, remove ambiguity, and post the versioned execution plan whose highest version is the plan.
spikeRun an exploratory, non-production experiment to decide whether an idea deserves a project. Reproduce current behaviour as a validated baseline, design a falsifiable experiment, measure gain and regression, produce inspectable evidence, and record a directional verdict plus a /sy:plan-ready brief wh
standardsResolve authoritative repository engineering policy for implementation, or run a cited conformance review. Loads only the mode-specific procedure and relevant rules.
tightenPROACTIVE — tighten one piece of already-written text before it is sent or posted, cutting padding without losing a fact. Route on destination: a console turn to the user, a plan's sign-off half, an issue body or PR description, a plan's /sy:ship half, a dispatch brief, or one of this plugin's own f
trackerOwn issue-tracker mechanics behind one contract, dispatching to the configured adapter (Jira or GitHub Projects): hierarchy, lifecycle, dependencies, comments, machine logs, attachments, and decomposition. Workflow skills own decisions; this skill owns durable representation and mutation.
whiteboardWork a fuzzy problem out with the user before anything is planned, measured or built. The user picks the surfaces: a decision page rendered from notes.md, a shared Excalidraw design board, or both. Every mode interviews the user hard on scope, requirements, constraints, resources and what done means
debateRuns one bounded proposer/adversary debate over the core decision behind a `/sy:plan` roadmap, `/sy:spec` plan, or `/sy:spike` verdict via `sy:debater`, and returns only the synthesized findings — never the raw exchange — for the user to steer.
debaterOne side of a proposer/adversary debate for `sy:debate`. Argues for or against a stated premise/approach using gathered evidence. Read-only.
explain-authorInvestigate one gnarly topic (a bug family, a design constraint, a system) for /sy:explain, verify every mechanism claim against source or a live repro, and author a layered, checkpointed explainer doc. Read-only outside the topic's resolved scratch directory.
gate-triageDisposition one gate round's findings for the /sy:ship GATE worker — accept with a fix spec, or reject with recorded reasoning — each keyed to a stable root cause. Source-read-only apart from its dispositions file; decides nothing else.
gateIndependent adversarial ship gate over one immutable base/head SHA pair. Review behaviour and standards, use hunt/refute for depth, and return a cited verdict.
huntDeep correctness investigation for gate. Hunt one coherent area for concrete bug candidates, or refute one candidate adversarially. Source-read-only; writes only in the repository's resolved scratch directory.
img-inspectorShort-lived visual inspector: receives image path(s) and an inspection task, looks at the pixels, and returns a text-only verdict. Keeps image tokens out of the long-running context.
repo-reviewRun the repository's own configured code-review skill over one pinned head SHA, verify its findings by reading the spans they cite, and return them as candidates for sy:gate to refute. Dispatches nothing. Never fixes, promotes, or dispositions.
repo-standardsRun the repository's engineering-standards pass — resolve mode for an implementation contract, review mode for a cited conformance review — and return only that skill's compact contract, keeping the rule reads out of the caller's context.
seamDepth investigation of one unclear architectural boundary for /sy:plan, read-only apart from its report file. Find the thinnest interface, hidden coupling, and real dependency order.
ship-buildBUILD worker for /sy:ship: implement and integrate the plan via bounded `sy:slice` delegates, verify against acceptance criteria, open the draft PR, and return the build brief.
ship-gateGATE worker for /sy:ship: a lightweight convergence loop that delegates the frontier verdict to `sy:gate`, CI triage to `/sy:ci`, and review-thread reconciliation to `/sy:pr`, drives the fix cycle, and promotes.
ship-startSTART worker for /sy:ship: read the plan file the parent materialised, delegate standards resolution, branch and worktree, seed resume state, and return the state brief.
sliceImplement one tightly specified, low-ambiguity /sy:ship slice in a caller-owned worktree. Test, commit locally, and return the commit plus dense verification pointers.
spec-gateIndependent pre-sign-off review of one fully drafted /sy:spec plan against the six-axis spec-gate checklist. Reports plan defects, never re-argues the core decision the debate pass already settled. Read-only apart from its report file.
sweepRead-only breadth reconnaissance. Map large code, PR, ticket, docs, or CI surfaces into pointer-dense leads for caller verification. Not for depth or verdicts.
traceDepth trace of one load-bearing behaviour or data path for /sy:spec, /sy:spike, or a /sy:ship parent, read-only apart from its report file. Follow it end to end, expose breaking cases, and return decisive evidence pointers.
syruns: pixi run --manifest-path ${CLAUDE_PLUGIN_ROOT}/pyproject.toml sy-server
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.guards.review_guard
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.guards.spec_gate_cap_guard
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.eval_events hook
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.usage hook
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.eval_events hook
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.usage hook
PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python -m sy_tools.eval_events hook
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
Use Alibaba Open Code Review from local Codex.
2 skills
Backend API design, GraphQL architecture, workflow orchestration with Temporal, and test-driven backend development
9 skills · 1 command · 8 agents
AI-supervised issue tracker for coding workflows
1 skill · 1 agent · hooks
PRD generation and conversion skills for the Ralph autonomous agent loop
2 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.