GREEN:

# Front Door Policy Step 2 Report 2026-05-18

## Purpose

内川様の「勝手にやってください」を、前段のStep 2許可として扱い、active runtime設定を変えずにFront Doorの古い全面禁止文言をcapability / risk-tier境界へ整理した。

## Changed Files

- `AGENTS.md`
- `docs/ai_workflow/00_FRONT_DOOR.md`
- `docs/ai_workflow/codex_hooks_setup_guide.md`
- `秘書/prompts/codex_task_frontdoor_template.md`
- `scripts/ai_workflow/frontdoor_check.py`
- `scripts/secretary/frontdoor_design_reform_lint.py`

## What Changed

- `AGENTS.md` の `Keep hooks disabled` を、旧Claude hook stack無効維持とCodex project-local check-only hooksの承認制pilotに分離した。
- `AGENTS.md` の `MCP integration / Codex Skill conversion / external API execution / UI automation はMVP外` という古い全面禁止を、read-only / check-only / prepare-before-save と write/final-action の境界へ置換した。
- `docs/ai_workflow/00_FRONT_DOOR.md` のMVP説明を、blanket feature banではなくcapability-based boundaryへ変更した。
- `docs/ai_workflow/codex_hooks_setup_guide.md` にStage 0からStage 3のcheck-only hook pilot pathを追加した。
- `秘書/prompts/codex_task_frontdoor_template.md` の通常タスク禁止文を、通常タスク限定に狭めた。Front Door / UWCP / Evidence / hooks / Codex governance改善が明示された場合は、active runtime設定を変えないローカルdocs/scripts/tests/report作成を許可する形へ整理した。
- `frontdoor_check.py` と `frontdoor_design_reform_lint.py` は、旧文言を必須/検出対象にする状態から、新しい境界文言を正常系として扱う状態へ更新した。

## Verification

- `python -m py_compile scripts\secretary\frontdoor_design_reform_lint.py scripts\ai_workflow\frontdoor_check.py`: PASS
- `python scripts\secretary\frontdoor_design_reform_lint.py --write-json tmp\frontdoor_policy_step2_after_lint_20260518.json`: PASS, findings_count 0
- `python scripts\ai_workflow\frontdoor_check.py`: GREEN
- `python scripts\ai_workflow\evidence_check.py --sample`: GREEN
- `python -m unittest tests.test_universal_work_control_plane`: PASS, 24 tests
- `python scripts\secretary\evidence_package_lint.py tmp\frontdoor_policy_step2_evidence_20260518.json`: PASS
- `python scripts\secretary\uwcp_adoption_gate.py --intent tmp\frontdoor_policy_step2_intent_20260518.json --evidence tmp\frontdoor_policy_step2_evidence_20260518.json --write-receipt tmp\frontdoor_policy_step2_adoption_receipt_20260518.json`: ADOPTED

## Unexecuted High-Risk Operations

- hooks activation: not executed
- `.codex/hooks.json` change: not executed
- settings change: not executed
- MCP configuration change: not executed
- env/credential/secret change: not executed
- production read/write: not executed
- save/send/purchase/friend-add completion: not executed
- commit/push/PR/merge: not executed

## PR / Commit State

- commit: not executed
- push: not executed
- PR: not created
- merge: not executed

## Decision Point

No ChatGPT decision is required for Step 2. The next separate decision is whether to run Step 3: a Codex project-local check-only hooks pilot. Step 3 would be a separate approval boundary because it may touch `.codex/hooks.json` or runtime hook activation.