implement
Builds finished project work with tests and a final check to make sure everything works.
Installation
Paste this into Claude Code, Cursor, or any agent that can run commands.
SKILL.mdShow the author's original SKILL.md
--- name: implement description: Implement approved project work with tests and final review. disable-model-invocation: true --- # Implement Implement approved work described by the active project's spec or tickets. Resolve every `.agents/projects/` path from the repository root; never read or write project artifacts in a global agent-installation directory. ## Process 1. Identify the active project. Read `.agents/projects/<project>/CONTEXT.md`, `LINKS.md`, `todo.md`, the approved spec, ticket drafts, tracker tickets, and relevant ADRs. Confirm the implementation target and its acceptance criteria before changing code. Done when the approved scope, test seams, and current ticket frontier are known. 2. Implement one vertical slice at a time. Use `tdd` at the pre-agreed seams where possible. Run typechecking and focused tests regularly; keep the project work list current as slices land. Done when each accepted slice is implemented and its focused verification passes. 3. Run the full relevant test suite and the repository's coverage tooling. Target 100% coverage for changed code and require at least 95%, unless the user or the repository explicitly opts out. Close meaningful gaps through behavior-focused tests, not coverage-only assertions. Done when verification passes and coverage meets the threshold, or the explicit exception is recorded in `.agents/projects/<project>/todo.md`. 4. Update project artifacts with completed work, material facts, decisions, and ticket links. If posting a Jira or GitHub comment, end it with the exact line: ```text Written by Cursor ``` Done when the project state accurately reflects the implementation result. 5. Invoke `code-review` against the implementation's fixed point and originating local spec or tracker ticket. Address confirmed findings, then rerun affected verification. Done when the review report is recorded and remaining findings are either fixed or explicitly reported to the user.
Mirrored from the author's public source. Install counts from the open skills registry.