Verification Before Completion
Jesse VincentRequires the agent to run verification commands and confirm the output before it claims that work is complete or passing.
By agent · Codex
Skills in the open SKILL.md format that Codex reads. Each one has a plain-language summary and the command to install it.
$ npx skills add <author>/<skill> --agent codexReplace the placeholder with the author and skill name from the skill's page. Start a new Codex session afterwards so the skill is picked up.
Requires the agent to run verification commands and confirm the output before it claims that work is complete or passing.
Writes a step-by-step implementation plan from a spec or requirements before any code is touched.
Guides creating, editing and testing skills before they are deployed.
Tells you which of Matt Pocock's skills or flows fits your situation.
Reviews the changes since a commit or branch on two axes: the repo's coding standards and the originating spec.
Gives the agent a shared vocabulary for designing deep modules, deciding interfaces and making code easier to test.
Runs a diagnosis loop for hard bugs and performance regressions.
Builds and sharpens a project's domain model through a glossary and ADRs.
Interviews you relentlessly to sharpen a plan or design, and writes ADRs and a glossary along the way.
Implements in code what the to-spec and to-tickets skills produced.
Implements a piece of work in code, working from a spec or a set of tickets.
Scans a codebase for chances to deepen modules, shows them in an HTML report, then interviews you about the one you pick.
Writes the body of a pull request so that reviewers can follow the change.
Builds a throwaway prototype to answer a design question about a state model, logic or UI.
Investigates a question against primary sources and saves the findings as a Markdown file in the repo.
Runs a retrospective on a coding session to capture what went well and what did not.
Configures a repo for Matt Pocock's engineering skills: issue tracker, triage labels and domain doc layout.
Builds features and fixes bugs test-first with the red-green-refactor loop.
Turns the current conversation into a spec and publishes it to the project's issue tracker, without a further interview.
Breaks a plan or spec into tracer-bullet tickets with their blocking relationships, published to the issue tracker.
Moves issues and external pull requests through triage: categorise, verify and write briefs an agent can act on.
Plans work too large for one agent session as a map of decision tickets, then resolves them one at a time.
Generates an interactive bash wizard that walks a person through steps only they can perform, such as credentials or dashboards.
Migrates test files from `as` type assertions to @total-typescript/shoehorn.
Questions
Often. A skill is a folder of instructions and scripts in the open SKILL.md format, which Codex reads. A skill that depends on tools only one agent provides will not carry over, so read the skill's source before you rely on it.
Each skill page has an install command. Run it in a terminal, then start a new Codex session so the skill is picked up.
An instructions file is loaded in every session and describes the project. A skill is loaded only when a task calls for it, so it can be long and specific without costing context on every turn.