# Implement

> Executes implement-stage loop workflow for the next scoped work item through stage-chain orchestration. Use when implement stage enters loop mode.

- Skill: `majiayu000/implement-18` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add majiayu000/implement-18`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/implement-18/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/implement-18

---


Follow `feature-dev-aidd:aidd-core` and `feature-dev-aidd:aidd-loop`.

## Steps
1. Inputs: resolve active `<ticket>/<scope_key>` and confirm loop artifacts are present for implement stage.
2. Stage-chain policy: execute via canonical stage-chain orchestration through stage runtime entrypoint; internal preflight/postflight are orchestration details and not operator commands.
3. Manual write/create of `stage.implement.result.json` is forbidden; stage-result files are produced only by stage-chain postflight. `[AIDD_LOOP_POLICY:MANUAL_STAGE_RESULT_FORBIDDEN]`
4. Runtime-path safety: use only runtime commands from this skill contracts. If stdout/stderr contains `can't open file .../skills/.../runtime/...`, stop with immediate BLOCKED `runtime_path_missing_or_drift`; one runtime-path error is terminal for current run, do not invent alternate filenames, and do not retry guessed commands.
5. Retry safety: do not rerun the same failing shell command more than once without new evidence/artifacts. For cwd/build mismatches, stop with blocker and handoff instead of looped retries.
6. Read order after stage-chain preflight artifacts: `readmap.md` -> loop pack -> review pack (if exists) -> rolling context pack; do not perform broad repo scan before these artifacts.
7. Run subagent `feature-dev-aidd:implementer`.
8. Orchestration: use the existing rolling context pack (do not regenerate it), Fill actions.json (v1) at `aidd/reports/actions/<ticket>/<scope_key>/implement.actions.json`, keep action types strictly in `{tasklist_ops.set_iteration_done, tasklist_ops.append_progress_log, tasklist_ops.next3_recompute, context_pack_ops.context_pack_update}`, and validate schema via `python3 ${CLAUDE_PLUGIN_ROOT}/skills/implement/runtime/implement_run.py`.
9. Canonical stage-chain: internal preflight -> stage runtime -> actions_apply.py/postflight -> `python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/stage_result.py`; it must produce `aidd/reports/loops/<ticket>/<scope_key>/stage.implement.result.json`. `[AIDD_LOOP_POLICY:CANONICAL_STAGE_RESULT_PATH]`
10. Non-canonical stage-result path under `skills/aidd-loop/runtime/` is forbidden (treat as prompt-flow drift). `[AIDD_LOOP_POLICY:NON_CANONICAL_STAGE_RESULT_FORBIDDEN]`
11. Output: return stage contract + updated artifacts with explicit handoff/next action.

## Command contracts
### `python3 ${CLAUDE_PLUGIN_ROOT}/skills/implement/runtime/implement_run.py`
- When to run: as canonical implement stage runtime before postflight.
- Inputs: ticket, scope/work-item context, and validated actions payload.
- Outputs: stage validation artifacts and status payload for downstream postflight.
- Failure mode: non-zero exit for invalid actions schema or missing stage prerequisites.
- Next action: fix actions/preconditions and rerun runtime validation before postflight.

### `python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-docio/runtime/actions_apply.py`
- When to run: mandatory final step in stage-chain postflight after actions validation.
- Inputs: `--actions <path>` and optional `--apply-log <path>`.
- Outputs: applied actions, progress/status artifacts, and apply logs.
- Failure mode: DocOps apply failure, boundary guard failure, or status-summary failure.
- Next action: inspect action/apply logs, fix root cause, rerun the stage-chain, and verify canonical stage result exists (`aidd/reports/loops/<ticket>/<scope_key>/stage.implement.result.json`).

### `python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/stage_result.py`
- When to run: stage-chain postflight stage-result emission only (not operator/manual recovery command).
- Inputs: canonical postflight payload (`ticket`, `stage`, `result`, `scope-key`, `work-item-key`, evidence links).
- Outputs: canonical `aidd.stage_result.v1` at `aidd/reports/loops/<ticket>/<scope_key>/stage.implement.result.json`.
- Failure mode: non-zero exit on missing required args or invalid stage-result contract fields.
- Next action: fix postflight payload generation and rerun the stage-chain; do not switch to non-canonical loop runtime paths.

## Notes
- Implement stage does not run ad-hoc test loops; format-only is allowed and test orchestration is delegated to hook policy.

## Additional resources
- Contract schema: [CONTRACT.yaml](CONTRACT.yaml) (when: preflight/postflight or actions contract is unclear; why: validate required fields and artifact expectations before rerun).