From eng
Final pass of the eng pipeline. Implements a fully-planned .plan ticket: writes the unit tests the doc specified (TDD — failing first), implements against the acceptance criteria following the repo's conventions, runs the tests until they pass, then verifies the acceptance criteria and walks the manual QA checklist. Updates the doc's Work Log and the README status. Use when a tested .plan doc is ready to build. Triggers: "execute", "implement this", "build this ticket", "do the work".
How this skill is triggered — by the user, by Claude, or both
Slash command
/eng:executeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are the final pass. The doc is scoped, grilled, reviewed, and has a test
You are the final pass. The doc is scoped, grilled, reviewed, and has a test contract. Your job is to build it the way the plan says and prove it works.
First, read the doc schema at ${CLAUDE_PLUGIN_ROOT}/SPEC.md and the target
ticket doc end to end — Acceptance Criteria, Test Plan, Codebase Touchpoints,
and Conventions & Patterns are your instructions.
If multiple tickets, choose the lowest-numbered unimplemented one whose
depends-on are all implemented. Don't start a ticket whose dependencies
aren't done. Confirm with the user before writing code (this is the gate — never
auto-run implementation silently). Create/checkout a branch and record it in the
doc frontmatter branch: and the README table.
Implement the ### Unit Tests spec — following the existing test files/patterns
it cites. Run them; they should fail for the right reason. This locks the
contract before you write the implementation.
Build the change using the ## Codebase Touchpoints as your map and the
## Conventions & Patterns as your style guide — match how this repo already
builds APIs/components/etc. so the code looks like the team wrote it. Stay inside
## Scope (respect the Out list — no creep).
Run the unit tests until green. Fix the implementation, not the tests (unless a test was wrong — note that in the Work Log).
Go through ## Acceptance Criteria one by one and check each off only when
actually satisfied. If one can't be met, stop and surface it — don't fake it.
Run the ### Manual / UI QA steps yourself where you can (use a browser/QA
skill if available); for steps that need a human, present the checklist and ask
the user to confirm. Record outcomes.
## Work Log: what you implemented, plan-vs-reality deviations,
test results, QA outcomes.phase: implemented; update the README table (phase + branch)./ship or the project's PR flow if they want to land it.Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub nitintf/ai-skills --plugin eng