Skip to main content

Using Matt Pocock Skills with Codex

The Matt Pocock skills give Codex a structured way to turn an idea into a reviewed change: clarify the problem, make a spec, split the work, implement it, and verify it. They are available globally to Codex in this workspace. Use this page to choose a skill by the problem in front of you. Use the generated skills reference when you need the exact source instructions for a skill.
This guide does not replace this repository’s operating rules. AGENTS.md, the task done gate, the Linear-Codex process, and screenshot-based verification take precedence over any upstream skill instruction.

Start here

Choose the smallest entry point that matches the work: Do not run every skill for every task. A small, known-path documentation or copy edit should still use the repository’s micro-change fast path. For normal implementation and verification work in this repository, begin with the local qa-task workflow and use the skills below when their specific problem applies.

The primary flow

handoff is the bridge when the conversation needs to move to a fresh Codex task. prototype is a temporary detour when a runnable experiment will answer a question more reliably than discussion.

Planning and discovery skills

Building and quality skills

Context, language, and learning skills

Intake and triage skills

For this repository, do not rerun setup-matt-pocock-skills to select a tracker without resolving the existing tracker-policy mismatch first. Full Codex tasks use Linear as the ownership source of truth, even where older documentation mentions GitHub Issues.

Repository-specific rules that always win

  1. Tracking: Full-process work must have a Linear issue, task file, handoff, and milestone comments. Do not publish work tickets to GitHub merely because an upstream skill uses GitHub examples.
  2. Verification: User-visible work needs actual end-to-end proof and a screenshot on the canonical route or artifact before it can be described as complete.
  3. Debugging: Observe the running system and identify the first failing boundary before changing product code. diagnosing-bugs complements this requirement; it does not replace it.
  4. UI work: Reuse the canonical shared route, primitive, or pattern first, and apply the Alleato product noise gate.
  5. Publishing: The main application and this ReadMe site have separate Git boundaries. Commit and publish only the files owned by the task in the repository that owns them.

Examples

A new feature that affects several surfaces

Use grill-with-docs to settle the outcome. Use to-spec to capture the accepted behavior, then to-tickets to create vertical slices. Run implement for one slice at a time, followed by code-review and the repository’s browser and screenshot verification gates.

A production bug with an unclear cause

Start with diagnosing-bugs, using browser/runtime evidence to establish a repeatable failure. Once the first broken boundary is known, use tdd or implement for the repair and keep the regression test.

An ambiguous architectural improvement

Use improve-codebase-architecture to find a high-leverage opportunity. Use codebase-design when choosing the seam, then return to the normal planning and implementation flow only after the problem is concrete.

See also