grill-me: Get Grilled by AI Before It Starts Coding
Walks through the most popular skill in the repo, the one-question-at-a-time interrogation that forces alignment before an agent writes a line of code.
01Coder on YouTube
A focused daily engineering loop built around alignment, domain language, feedback, and codebase design.
Matt’s collection is an expert’s personal toolbox rather than a framework that owns the whole development process. It keeps the engineer in control, treats context as scarce, and now draws a sharper line between user-invoked orchestration and model-invoked disciplines. The emerging loop persists decisions in specs and tickets, schedules implementation from a dependency graph, reviews from fresh context, and uses retrospectives to improve the agent environment after the work.
npx skills@latest add mattpocock/skillsExperienced individual developers and small teams that want a sharp daily workflow, especially for requirements interrogation, domain modeling, TDD, debugging, deep module boundaries, and explicit human control over when larger workflows begin.
The 25 promoted skills remain strongest as a Claude Code-centered personal workflow, although editable copies work across agents. The most ambitious orchestration and self-improvement ideas are still under in-progress, rely on background agents, worktrees, issue graphs, and project conventions, and should not yet be treated as stable lifecycle coverage.
Latest evolution
Reviewed 2026-09-12 · upstream 3cca18b
The promoted catalog is still a deliberately small set of 25 engineering and productivity skills. The new direction is deeper orchestration rather than a larger stable menu: implementation is becoming a dependency-graph scheduler for parallel agents, while retrospective analysis turns failures from one session into better navigation, checks, review rules, tool use, and information access for the next.
The repository now defines user-invoked skills as orchestration entry points and model-invoked skills as reusable disciplines, preventing one command layer from silently chaining into another.
Claude Code users can subscribe to a managed official-marketplace plugin, while the skills CLI remains the editable, cross-agent path; installing both would duplicate every skill.
The local linking script now excludes misc utilities as well as deprecated skills, confirming that the daily-driver collection is narrower than every public SKILL.md in the repository.
External reviewers consistently value the grilling, durable context, and human control, while also noting lighter launch coverage, weaker default delegation, and extra friction on large or already-clear tasks.
Treats tickets as a dependency graph, runs ready work in parallel worktrees, merges completed slices, and closes with independent review.
Reviews a coding session for durable improvements to navigation, automated checks, standards, tool economy, instructions, and information access.
Start here
These are the skills with the clearest standalone value. Each review explains the benefit and the cost.
The clearest reason to install Matt’s collection. Grill Me turns vague intent into a visible decision tree and refuses to let the agent silently fill gaps, which directly addresses the most expensive failure in agentic coding: building the wrong thing quickly.
Grill With Docs improves on a normal requirements interview by turning resolved language and architectural choices into durable project context. That makes later sessions shorter and reduces inconsistent naming across both code and agent explanations.
A pragmatic, seam-oriented TDD guide that insists on observing the failure before implementation and favors tests through public behavior. It is especially useful when coding agents otherwise add tests after the fact to bless what they already wrote.
This skill makes debugging a sequence of evidence-producing gates: reproduce, minimize, hypothesize, instrument, fix, and regression-test. It is a strong antidote to agents editing the first suspicious line and declaring victory.
Codebase Design gives agents a compact architectural vocabulary centered on deep modules, narrow interfaces, locality, and testable seams. It improves design conversations without pretending that a generic architecture pattern fits every repository.
Watch & read
4 videos · 1 post covering this collection. Videos play right here.
Walks through the most popular skill in the repo, the one-question-at-a-time interrogation that forces alignment before an agent writes a line of code.
01Coder on YouTube
Second part of the engineering-chain series, showing how to-spec converts an open-ended discussion into a document the team can actually sign off on.
01Coder on YouTube
Third part of the engineering-chain series, demonstrating how to-tickets slices an approved spec into ordered, independently verifiable work.
01Coder on YouTube
Matt Pocock on the mistakes people make writing their own skills, and why fewer, sharper skills beat a large unfocused catalog.
01Coder on YouTube
Stable inventory
In-progress experiments are excluded from this editorial collection.
Reviews a diff independently against repository standards and the originating specification.
View skill sourceDesigns deep modules with small interfaces, clean seams, and testable behavior.
View skill sourceSharpens project terminology, edge cases, context documents, and architectural decisions.
View skill sourceSurveys a codebase for high-leverage opportunities to deepen module boundaries.
View skill sourceRuns a gated diagnosis loop from reproduction and minimization through instrumentation and regression testing.
View skill sourceInterrogates a design while building shared vocabulary and recording decisions as it goes.
View skill sourceRelentlessly interviews the user until every reachable branch of a decision is resolved.
View skill sourceProvides the reusable design-tree interrogation discipline behind Matt’s interview workflows.
View skill sourceProduces a focused questionnaire for a domain expert who holds missing decisions.
View skill sourceImplements an approved specification or ticket set through explicit engineering seams.
View skill sourceBuilds disposable code to answer a focused state, logic, or interface design question.
View skill sourceInvestigates a technical question against primary sources and saves a cited research artifact.
View skill sourceResolves conflicts hunk by hunk from the documented intent of both branches.
View skill sourceConfigures issue tracking, triage vocabulary, and project documentation conventions.
View skill sourceApplies a strict red-green loop and defines what makes an integration test worth keeping.
View skill sourceSynthesizes the current conversation into a specification without restarting discovery.
View skill sourceBreaks work into tracer-bullet tickets with explicit blocking relationships.
View skill sourceMoves issues and external pull requests through defined verification and briefing states.
View skill sourceMaps large multi-session work as a graph of decisions before implementation begins.
View skill sourceCreates an interactive script for manual infrastructure, credentials, and cutover steps.
View skill sourceCompacts a conversation into a durable artifact that another agent can continue from.
View skill sourceRuns a stateful, multi-session learning workflow inside a dedicated workspace.
View skill sourceMakes skills and instruction documents produce predictable agent behavior.
View skill sourceExplore the other expert toolkits by workflow, scope, platform, and strongest use cases.