← ClaudeAtlas

implementing-sliceslisted

Slice-by-slice execution procedure for the implementer agent — dispatch modes (initial and review-fix), the slice-execution loop, TDD discipline, blocker handling, and the scope fence. Loaded when an implementation plan is executed or a hard-gate review failure needs fixing.
bostonaholic/team · ★ 11 · AI & Automation · score 75
Install: claude install-skill bostonaholic/team
# Implementing Slices The implementer's execution procedure: consume the plan, work through one vertical slice at a time, and commit each slice atomically the moment its tests pass. ## Dispatch modes The orchestrator dispatches you with the artifact directory `docs/plans/<id>/`. ### Initial dispatch (after the test-architect's failing tests are confirmed) Your inputs are the plan (`docs/plans/<id>/plan.md` — slice list, file-level steps, per-slice tests), the structure (`docs/plans/<id>/structure.md` — order and verification checkpoints), the failing acceptance tests (the completion contract), and `docs/plans/<id>/repos.md` when present. `repos.md` defines multi-repo mode: each repo's slug, absolute path, and worktree path (under `## Worktrees`); every plan step annotated `[repo: <slug>]` is applied inside that repo's worktree — `cd` there before running the step's edits, tests, and commits. Before implementing, run the test suite once (in each involved worktree, in multi-repo mode) to establish the baseline of failing tests. ### Review-fix dispatch (after a hard-gate failure) When dispatched after the aggregate gate fails, you are in a **fix loop**. The orchestrator passes you a typed failure class and the reviewers' findings. Fix what the findings name, under these constraints: - **Security findings: never weaken the fix.** Fix each vulnerability directly — parameterize the query, remove the hardcoded secret, add the auth check, escape the output. The security r