--- name: execute description: Execute an approved plan step by step, verifying each step before proceeding. argument-hint: " [--step N] [--continue]" --- You are in EXECUTION mode. Follow the plan precisely — do not improvise. Input: slug = $0 Options parsed from $ARGUMENTS: - `--step N` — start from step N (skip earlier steps, assume already done) - `--continue` — resume from the first step not yet logged as done in the Decision log Target: `.claude\/plans\/.md` ## 1. Load and Validate Read `.claude\/plans\/.md`. **Gate checks — stop if any fail:** - Status MUST be `Approved`. If `Draft` → tell user to run `\/approve ` first. If `Completed` → tell user plan is already done. - No blocking Open questions. Read `CLAUDE.md` at repo root if it exists — follow project conventions during implementation. ## 2. Determine Starting Step - Default: Step 1 - `--step N`: Start at step N - `--continue`: Scan the Decision log for entries like `Step N — done` or `Step N — verified`. Start at the first step without such an entry. Print: ``` Executing: .claude\/plans\/.md Starting at: Step N of M ``` ## 3. Execute Each Step For each step in order: ### 3a. Announce ``` ━━━ Step N\/M: ━━━ ``` ### 3b. Implement - Read the **Change** description - Implement exactly what it says — no more, no less - Follow project patterns from CLAUDE.md - If the step is ambiguous, stop and ask the user rather than guessing ### 3c. Verify - Run the **Verify** command or check described in the step - If verification **passes**: log ` — Step N — verified ✓` in the Decision log - If verification **fails**: 1. Attempt ONE fix based on the error output 2. Re-run verify 3. If it fails again → stop execution, log ` — Step N — BLOCKED: `, update Status to `Blocked`, and print: ``` BLOCKED at Step N: Reason: Options: \/plancheck — review and adjust the plan \/execute --continue — resume after fixing manually ``` ### 3d. Checkpoint After each verified step, check: - Did the change introduce something unexpected (new test failures, lint errors)? - If yes → stop and report before continuing ## 4. Run Test Plan After all steps pass: - Execute the **Test plan** commands from the plan - If tests pass: log ` — Test plan — passed ✓` - If tests fail: log ` — Test plan — FAILED:

`, do NOT mark as completed ## 5. Check Acceptance Criteria Review each acceptance criterion in the plan: - Check the box `[x]` if satisfied - If any criterion is NOT met, list what's missing and stop ## 6. Complete When all steps verified + test plan passed + acceptance criteria met: - Set `Status: Approved` to `Status: Completed` - Log ` — Plan completed` - Print: ``` Plan completed — .claude\/plans\/.md Steps: N\/N verified Test plan: passed Acceptance criteria: all met ``` ## Rules - Follow the plan exactly. Do not add features, refactor surrounding code, or "improve" things not in the plan. - If reality diverges from the plan (e.g., a file doesn't exist, an API changed), STOP and tell the user. Suggest running `\/plancheck ` to update the plan. - One step at a time. Do not batch or skip steps. - Every code change must be tied to a specific step in the plan. - Update the Decision log as you go — it's the audit trail. - Do not combine the execute skill with other skills in the same invocation.