# Agentic Spec Confirmed Implementation

> Orchestrate a Phase 6.5 confirmed implementation request from either a fenced JSON spec-confirmed-implementation-request or an explicit natural-language automation request after request assembly and confirmation; bare acknowledgements do not trigger this skill.

- Skill: `dolphinai2026/agentic-spec-confirmed-implementation` (Agent Skill, multi-file: 15 files)
- Install (CLI): `npx skillmds@latest add dolphinai2026/agentic-spec-confirmed-implementation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dolphinai2026/agentic-spec-confirmed-implementation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dolphinai2026 (https://skillmd.com/u/dolphinai2026)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dolphinai2026/agentic-spec-confirmed-implementation

---


# Agentic Spec Confirmed Implementation

Use `rules/phase65-trigger.md` for the complete trigger and non-trigger
decision. This skill runs Path A fenced JSON requests directly after
validation, and runs Path B only after the Agent has assembled the full request
JSON, shown it to the user, received explicit confirmation, validated it, and
passed the final same-phase manifest recheck.

## Required Context

- `rules/phase65-trigger.md`
- `rules/request.md`
- `rules/planning.md`
- `rules/execution.md`
- `rules/review-gap-policy.md`
- `rules/remote-sync-ci.md`
- `rules/manual-doc-handoff.md`
- `docs/solutions/l3/detail/phase6-5a-spec-confirmed-implementation-automation-graphify-chain-analysis.md`

## Workflow

1. Prepare (`prepare`): use validated request JSON as the input, with Path A
   fenced JSON and Path B assembled JSON as the upstream sources; validate the
   request and manifest, run `agentic-spec doctor`, execute confirm guard,
   confirm exactly one plan, confirm `codex-agentic` P0 detached support, and
   resolve `ExecutionRefs.phase_id`, plan refs, confirm commit, and phase refs
   before any executor start.
2. Start detached (`start detached`): run:

   ```text
   codex-agentic --automation phase-confirmed --detach --result docs/assets/builder/executions/<phase-id>.codex-agentic-run.yaml <plan_ref>
   ```

   Capture and persist the execution handle JSON emitted by the detached start.
   The handle, not terminal prose, is the control-plane correlation object.
3. Monitor/closeout (`monitor/closeout`): read detached lifecycle evidence from
   the normalized execution store and the phase run-result. Do not infer machine
   state, merge readiness, review status, push completion, cleanup completion, or
   CI handoff from stdout prose, HUD text, or terminal summaries.
4. Blocked intervention (`blocked intervention`): when lifecycle status is
   blocked, report the blocked reason code, safe evidence refs, and the
   lifecycle or prompt action type required for the next human decision.

Compatibility fallback for pre-S3 environments only: if detached P0 support is
unavailable and the confirmed request explicitly allows legacy operation, the
foreground command remains:

```text
codex-agentic --automation phase-confirmed --legacy-foreground --result docs/assets/builder/executions/<phase-id>.codex-agentic-run.yaml <plan_ref>
```

Foreground stdout is never an automation decision contract.

## Guardrails

- Do not ask for a second plan approval.
- Do not parse human-readable `codex-agentic` stdout or HUD text as the decision contract.
- Do not force push.
- Do not auto-patch app workspace `AGENTS.md` or `docs/solutions`.
- Do not put semantic decisions into `agentic-spec`.