Implements pre-planned tasks while enforcing project gates.

Install

mkdir -p .claude/skills/work && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/17563" && unzip -o skill.zip -d .claude/skills/work && rm skill.zip

Installs to .claude/skills/work

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Execute an approved plan step-by-step, with the repo's gate between steps and no scope creep. Use when implementing a planned change, working a checklist, or the user says "go" / "implement it" / "do the plan". Stops and reports on the first red gate.
251 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Intermediate

Key capabilities

  • Execute approved plans step-by-step
  • Ensure changes are green and reviewable after each step
  • Prevent scope creep during implementation
  • Match surrounding code style and delete dead code
  • Include tests as part of each step
  • Report on the first red gate encountered

How it works

The skill implements one step of an approved plan at a time, building it in the surrounding style and running the repo's gate. It stops if a gate fails, preventing further work on a broken step.

Inputs & outputs

You give it
work plan
You get back
Green, reviewable change for a single step

When to use work

  • Implementing an approved feature
  • Working through a development checklist
  • Executing a planned refactor

About this skill

Work the plan

Implement one step at a time. The point is a green, reviewable change — not speed. If there's no plan yet and the change is non-trivial, run /spec first.

The loop (per step)

  1. State the step in one line so progress is visible.
  2. Build it in the surrounding style. If you're unsure a function, option, or flag exists — yours or a dependency's — /verify-api before you write it; don't guess. Obey AGENTS.md and match .agent/rules/.
  3. Gate before moving on — run the repo's exact gate (AGENTS.md → "The gate").
  4. Red gate → stop. Don't pile the next step on a broken one. Fix it, or report the blocker with the error and your read on it. Never edit a test to make a real failure pass.

Rules while working

  • No scope creep. Build the approved slice. A good idea that wasn't in the plan → the backlog (coop backlog add) as "later", don't build it now. If the plan turns out wrong, stop and re-plan with the user — don't silently redesign.
  • Readable, no bloat. Match the surrounding style. Delete dead code you pass. No speculative options or abstractions. Comments say why, not what.
  • Tests are part of the step, not a follow-up — including the failure/denial path.
  • Greenfield. Replacing code → delete the old and update every caller in the same change; no shims for behavior nothing depends on yet.

When the change is green and reviewable, hand back for review (or, in a /sweep run, self-review the diff, commit, and tick the task).

When not to use it

  • When there is no approved plan
  • When speed is prioritized over green, reviewable changes
  • When the user wants to redesign the plan silently

Limitations

  • Stops on the first red gate
  • Does not allow scope creep
  • Does not allow silent redesigns of the plan

How it compares

This skill enforces a strict, step-by-step implementation process with gate checks and scope control, ensuring quality and preventing issues from compounding, unlike an unguided implementation.

Compared to similar skills

work side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
work (this skill)022dNo flagsIntermediate
cursor-local-dev-loop310dReviewBeginner
codex-worker15moReviewAdvanced
agent-challenges15moNo flagsIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry