Skip to content

Skills index

All skills live in mattpocock/skills on GitHub. Each is a SKILL.md file that an AI coding agent follows when invoked. Headings below are the skill slugs; the slash command is usually the same name, and where a separate command exists it is noted. The “Used in” line under each skill is generated from the situation pages’ skills tables.

Pocock groups them into engineering and productivity skills, and describes one main flow through them: grill an idea, model the domain, prototype what is uncertain, write a spec, break it into tickets, implement test-first, review. Several skills are on-ramps or side branches of that flow. The invocation cheat sheet lists every example command in this guide in one place.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, It does exactly what was asked and it is still wrong, The bug that defies understanding, and whether to understand it at all, Knowing what not to build: the 10x engineer as context

Invoked with /grill-me. Interviews you relentlessly about a plan, decision, or idea, working a design tree in rounds until nothing is left silently assumed. Facts are the agent’s job to look up; decisions are yours. Used in this guide wherever a request has to be turned into a problem.

Used in: The obvious data model breaks on reality, One word, many meanings across departments, Growing from craftsman to meta-rational engineer

Grilling combined with domain-modeling: the interview also writes glossary entries and ADRs as terms and decisions crystallise.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, The obvious data model breaks on reality, One word, many meanings across departments, It does exactly what was asked and it is still wrong, The codebase is a big ball of mud, Growing from craftsman to meta-rational engineer

Actively builds the project’s domain model: challenges overloaded terms, invents edge-case scenarios, and writes the glossary (CONTEXT.md) and decisions (ADRs) down the moment they settle. Reading the glossary is not this skill; changing it is.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, One word, many meanings across departments, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, Growing from craftsman to meta-rational engineer

Turns a decision you cannot answer alone into a Markdown questionnaire for one named person who holds the knowledge you lack. The interview is about the send, not the subject: who it goes to and what you need back.

Used in: The obvious data model breaks on reality, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, The codebase is a big ball of mud, The bug that defies understanding, and whether to understand it at all

Investigates a question against primary sources in a background agent and writes the findings to a Markdown file in the repository, each claim cited.

Used in: Purposes keep changing and belong to the whole context, not the requester, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, It does exactly what was asked and it is still wrong, Knowing what not to build: the 10x engineer as context, Growing from craftsman to meta-rational engineer

Plans work too large for one session as a shared map of decision tickets on the issue tracker, then resolves them one at a time until the way to a named destination is clear.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, It does exactly what was asked and it is still wrong, Knowing what not to build: the 10x engineer as context

Builds throwaway code that answers a design question: whether a state model feels right, what a UI should look like, whether an approach is viable. Deliberately omits production testing and polish.

Used in: no situation page yet.

Generates a bash script that walks a human through steps only a human can do, such as provisioning infrastructure or setting secrets.

Used in: Nobody knows what the problem is: requirements arrive as solutions, You cannot know yet: spikes and prototypes, Planning too much or too little: shaping work at the right altitude, It does exactly what was asked and it is still wrong, Knowing what not to build: the 10x engineer as context

Turns the current conversation into a spec, with no further interview, and publishes it to the issue tracker labelled ready for an agent.

Used in: Planning too much or too little: shaping work at the right altitude, The codebase is a big ball of mud

Breaks a plan or spec into tracer-bullet tickets, each declaring what blocks it, published to the tracker.

Used in: Planning too much or too little: shaping work at the right altitude, The bug that defies understanding, and whether to understand it at all, Knowing what not to build: the 10x engineer as context

Moves issues through a small state machine of triage roles: categorise, verify, grill if needed, and write agent-ready briefs.

Used in: Nobody knows what the problem is: requirements arrive as solutions, The obvious data model breaks on reality, One word, many meanings across departments, You cannot know yet: spikes and prototypes, The codebase is a big ball of mud, The bug that defies understanding, and whether to understand it at all

The red-green-refactor loop, with a reference for what makes a test worth keeping and which anti-patterns to avoid. Reads CONTEXT.md so test names use the project’s vocabulary.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, The obvious data model breaks on reality, One word, many meanings across departments, You cannot know yet: spikes and prototypes, The codebase is a big ball of mud, Knowing what not to build: the 10x engineer as context, Growing from craftsman to meta-rational engineer

Implements a spec or set of tickets, using tdd at agreed seams, with typechecks and tests run along the way.

Used in: One word, many meanings across departments, The codebase is a big ball of mud

Shared vocabulary for designing deep modules: a lot of behaviour behind a small interface at a clean seam. Other skills borrow its language.

Used in: no situation page yet.

Resolves an in-progress merge or rebase by finding the intent behind each side and preserving both where possible.

Used in: Nobody knows what the problem is: requirements arrive as solutions, Purposes keep changing and belong to the whole context, not the requester, It does exactly what was asked and it is still wrong, The codebase is a big ball of mud, Knowing what not to build: the 10x engineer as context

Reviews the diff since a fixed point along two axes, standards conformance and spec conformance, in parallel sub-agents. It does not ask whether the spec was the right one; several pages in this guide add that question.

Used in: The codebase is a big ball of mud, The bug that defies understanding, and whether to understand it at all

A diagnosis loop for hard bugs and performance regressions: reproduce, hypothesise, instrument, isolate, and only then fix.

Used in: The codebase is a big ball of mud

Scans a codebase for opportunities to deepen shallow modules, presents them as a report, and grills through whichever one you pick.

Used in: Knowing what not to build: the 10x engineer as context

A router: asks which skill or flow fits your situation.

Used in: Growing from craftsman to meta-rational engineer

Teaches a skill or concept across sessions, keeping the state of your learning in the workspace.

Used in: It does exactly what was asked and it is still wrong, The bug that defies understanding, and whether to understand it at all, Growing from craftsman to meta-rational engineer

Compacts the current conversation into a document another agent can pick up, pointing at specs, ADRs, and issues rather than repeating them.

Used in: One word, many meanings across departments

Stops the agent and asks it to re-pitch its last message in plain, simplified language using the project’s vocabulary.

Used in: Growing from craftsman to meta-rational engineer

A reference for writing any document an agent consumes: skills, AGENTS.md, or docs reached by a pointer.

Used in: no situation page yet.

Scaffolds the per-repository configuration the other skills assume: issue tracker, triage labels, and where the domain docs live.