@ps-neko/nekowork 0.1.0-alpha.10 → 0.1.0-alpha.12

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (74) hide show
  1. package/AGENTS.md +1 -1
  2. package/CLAUDE.md +3 -3
  3. package/README.ko.md +30 -30
  4. package/README.md +75 -53
  5. package/WORKING-CONTEXT.md +14 -4
  6. package/agent.yaml +5 -4
  7. package/agents/codex-challenger.md +1 -1
  8. package/agents/codex-reviewer.md +1 -1
  9. package/commands/nekowork-full-cycle.md +29 -0
  10. package/docs/ADVANCED.md +16 -0
  11. package/docs/ALPHA-RECRUITMENT.md +157 -0
  12. package/docs/ARCHITECTURE.md +2 -2
  13. package/docs/AUDIT.md +1 -1
  14. package/docs/CATALOG-PACKS.md +1 -1
  15. package/docs/CHANGELOG.md +42 -5
  16. package/docs/CLI-STAGES.md +8 -0
  17. package/docs/DEMO.md +1 -1
  18. package/docs/FEEDBACK-TRIAGE.md +26 -2
  19. package/docs/GUIDED-MODE.md +91 -0
  20. package/docs/PORTING.md +1 -1
  21. package/docs/POST-RELEASE-CHECKLIST.md +167 -0
  22. package/docs/QUICKSTART.md +7 -0
  23. package/docs/README-1.0-hero-draft.md +164 -0
  24. package/docs/RELEASE-READINESS.md +1 -1
  25. package/docs/SCOPE-1.0.md +309 -0
  26. package/docs/SETUP.md +1 -1
  27. package/docs/VISION.md +167 -0
  28. package/docs/assets/demo-terminal.svg +1 -1
  29. package/docs/assets/hero.gif +0 -0
  30. package/docs/{CLI-UX-PHASE1A-PLAN.md → dev-log/2026-05-13-cli-ux-phase1a-plan.md} +2 -0
  31. package/docs/{CLI-UX-REDESIGN.md → dev-log/2026-05-13-cli-ux-redesign.md} +2 -0
  32. package/docs/examples/github-actions-verify-pr.yml +90 -0
  33. package/manifests/install-components.json +6 -6
  34. package/manifests/install-modules.json +3 -3
  35. package/package.json +7 -2
  36. package/scripts/benchmark/rules.js +179 -0
  37. package/scripts/build-cursor.js +1 -1
  38. package/scripts/ci/security-hardening.js +27 -4
  39. package/scripts/cli/commands/cockpit-command.js +388 -0
  40. package/scripts/cli.js +179 -17
  41. package/scripts/demo-review.js +2 -2
  42. package/scripts/lib/diff-parser.js +293 -0
  43. package/scripts/lib/project-detector.js +309 -0
  44. package/scripts/lib/router.js +1 -1
  45. package/scripts/lib/rules/_helpers.js +149 -0
  46. package/scripts/lib/rules/auto-apply-commit-push.js +94 -0
  47. package/scripts/lib/rules/hardcoded-credential.js +103 -0
  48. package/scripts/lib/rules/package-lockfile-risk.js +92 -0
  49. package/scripts/lib/rules/secret-fallback.js +237 -0
  50. package/scripts/lib/rules/test-or-security-disable.js +91 -0
  51. package/scripts/orchestrators/review.js +1 -1
  52. package/scripts/orchestrators/verify-pr.js +466 -0
  53. package/scripts/sync-claude-md.js +1 -1
  54. package/skills/{claude-led-codex-review → nekowork-full-cycle}/SKILL.md +3 -3
  55. package/skills/plan-eng-review/SKILL.md +1 -1
  56. package/skills/ralph/SKILL.md +1 -1
  57. package/skills/review/SKILL.md +2 -2
  58. package/skills/ship/SKILL.md +1 -1
  59. package/skills/verified-gate/SKILL.md +47 -0
  60. package/LICENSE +0 -21
  61. package/commands/claude-led-codex-review.md +0 -29
  62. package/docs/CODEMAPS/README.md +0 -15
  63. package/docs/CODEMAPS/agents.md +0 -22
  64. package/docs/CODEMAPS/bridge.md +0 -18
  65. package/docs/CODEMAPS/hooks.md +0 -28
  66. package/docs/CODEMAPS/manifests.md +0 -15
  67. package/docs/CODEMAPS/rules.md +0 -22
  68. package/docs/CODEMAPS/schemas.md +0 -22
  69. package/docs/CODEMAPS/scripts.md +0 -198
  70. package/docs/CODEMAPS/skills.md +0 -31
  71. package/docs/CODEMAPS/tests.md +0 -144
  72. package/docs/workflows-stash/README.md +0 -32
  73. package/docs/workflows-stash/harness-review.yml +0 -166
  74. package/docs/workflows-stash/harness-validate.yml +0 -98
package/AGENTS.md CHANGED
@@ -28,7 +28,7 @@ fact_forcing: true|false # PreToolUse 사실 조사 강제 여부
28
28
  - `commands/` 는 legacy slash-entry 호환 표면이다. 신규 추가 금지, 점진 마이그레이션.
29
29
  - `agents/` 는 페르소나 카탈로그다. 워크플로우는 `skills/` 에서 정의하고 에이전트는 `skills/` 가 호출한다.
30
30
 
31
- ## 7단계 풀사이클 (claude-led-codex-review)
31
+ ## 7단계 풀사이클 (nekowork-full-cycle)
32
32
 
33
33
  | 단계 | 담당 | 입력 | 출력 |
34
34
  |---|---|---|---|
package/CLAUDE.md CHANGED
@@ -8,13 +8,13 @@
8
8
 
9
9
  ## 자동 갱신 영역
10
10
 
11
- <!-- HARNESS:START version=0.1.0-alpha.10 -->
11
+ <!-- HARNESS:START version=0.1.0-alpha.12 -->
12
12
  <!-- 이 영역은 scripts/sync-claude-md.js 가 자동 갱신한다. 직접 편집 금지. -->
13
13
 
14
14
  ## 카탈로그 요약
15
15
 
16
16
  - agents: 11
17
- - skills: 10
17
+ - skills: 11
18
18
  - commands: 1 (legacy compat)
19
19
  - hooks: 5 (gateguard-fact-force, config-protection, quality-gate, pre-bash-dispatcher, persistent-mode)
20
20
  - packs: core, builder, productivity, team, debugging, maintenance, pr, catalog-plus, quality, security, frontend, testing, release, enterprise
@@ -93,7 +93,7 @@ nekowork costs --since=7d # 비용 추정
93
93
 
94
94
  ## 매직 키워드 → 스킬 (명시 옵트인만)
95
95
 
96
- 자동 활성 키워드 감지는 **사용**하지 않는다. 사용자 룰("확인 후 실행") 우선. 모든 스킬은 슬래시 명령(`/claude-led-codex-review`) 또는 CLI(`nekowork review`) 로 명시 호출.
96
+ 자동 활성 키워드 감지는 **사용**하지 않는다. 사용자 룰("확인 후 실행") 우선. 모든 스킬은 슬래시 명령(`/nekowork-full-cycle`) 또는 CLI(`nekowork review`) 로 명시 호출.
97
97
 
98
98
  ## 핸드오프 5필드
99
99
 
package/README.ko.md CHANGED
@@ -2,17 +2,17 @@
2
2
 
3
3
  [English](README.md) | [한국어](README.ko.md)
4
4
 
5
- AI 코드 변경을 위한 검증 기반 오토파일럿입니다.
5
+ **AI 만든 코드, 검증 없이는 통과시키지 마세요.**
6
6
 
7
7
  [![validate](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml/badge.svg)](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml)
8
8
 
9
- AI가 만들고, Codex가 검증하고, 사람은 최종 적용 경계를 승인합니다.
10
-
11
- NEKOWORK는 AI가 계획, 수정, 검증, 제한된 재수정, 리포트 생성을 수행하도록 돕습니다. 하지만 최종 `apply`는 항상 사람이 명시적으로 실행해야 합니다.
9
+ NEKOWORK 는 AI 생성한 코드를 위한 로컬 검증 게이트입니다. diff 를 분석하고, 결정적 위험 룰을 실행하고, 증거를 수집한 뒤, 머지 / 적용 가능 여부를 판정합니다 — auto-commit / auto-push 없이, LLM 판정에 의존하지 않고.
12
10
 
13
11
  > 이 문서는 한국어 요약본입니다. 전체 상세 설명과 모든 고급 옵션은 [English README](README.md)를 참고하세요.
14
12
 
15
- 여기서 "검증됨"은 정답을 수학적으로 보증한다는 뜻이 아닙니다. 독립 리뷰, 테스트 evidence, 위험 정책, Human Gate, 명시적 apply 경계를 기록했다는 뜻입니다.
13
+ 여기서 "검증됨"은 정답을 수학적으로 보증한다는 뜻이 아닙니다. verdict 결정적 룰과 검증 결과만 결정합니다. 선택적 Codex 리뷰는 advisor 노트로만 기록되며 verdict 에 영향을 주지 않습니다.
14
+
15
+ > 1.0 scope 와 로드맵: [docs/SCOPE-1.0.md](docs/SCOPE-1.0.md). 장기 비전 (검증 우선 AI 개발 공장): [docs/VISION.md](docs/VISION.md).
16
16
 
17
17
  ## 용어
18
18
 
@@ -25,12 +25,13 @@ NEKOWORK는 AI가 계획, 수정, 검증, 제한된 재수정, 리포트 생성
25
25
  ## 핵심 원칙
26
26
 
27
27
  ```text
28
- NEKOWORK = 검증 기반 오토파일럿 -> Codex 검증 -> Human Gate -> 명시적 apply
28
+ NEKOWORK = diff -> 결정적 위험 룰 -> 검증 명령 -> 증거 -> 결정적 verdict -> REPORT -> Human Gate -> 명시적 apply
29
29
  ```
30
30
 
31
31
  ```text
32
- apply 전까지는 자율적으로.
33
- ship 전에는 독립 검증.
32
+ 증거 없으면 통과 없음.
33
+ LLM 의견은 verdict 아님.
34
+ 테스트 없으면 PASS 아님 (INSUFFICIENT_EVIDENCE).
34
35
  경계에서는 사람이 통제.
35
36
  ```
36
37
 
@@ -50,42 +51,41 @@ NEKOWORK는 기본 흐름을 mock provider 모드로 확인할 수 있습니다.
50
51
 
51
52
  ## 30초 실행
52
53
 
53
- 현재 npm alpha를 바로 실행할 있습니다.
54
+ 요구사항: Node.js 22+, npm, git. commit 하나 이상 있는 git repo.
54
55
 
55
56
  ```bash
56
57
  npx -y @ps-neko/nekowork@alpha check
57
- npx -y @ps-neko/nekowork@alpha start "fix failing tests safely" --session first-start
58
- npx -y @ps-neko/nekowork@alpha report --session latest
58
+ npx -y @ps-neko/nekowork@alpha verify-pr
59
+ cat REPORT.md
60
+ cat .nekowork/decision.json
59
61
  ```
60
62
 
61
- 먼저 실행 경로만 보고 싶다면:
63
+ `check` 환경을 진단합니다. `verify-pr` 가 현재 working tree diff 를 결정적 위험 룰로 스캔하고, `.nekowork/evidence/` 에 증거를 남기고, 머지/적용 가능 여부를 판정합니다. 프로젝트 루트에 `REPORT.md` 와 `.nekowork/decision.json` 을 작성합니다.
62
64
 
63
- ```bash
64
- npx -y @ps-neko/nekowork@alpha start "fix failing tests safely" --dry-run
65
- npx -y @ps-neko/nekowork@alpha auto "refactor this safely" --parallel-candidates 2 --dry-run
66
- ```
65
+ > **재현성 메모:** `npx @ps-neko/nekowork@alpha` 는 가장 최근 publish 된 alpha 로 resolve 됩니다. publish 된 alpha 는 `main` 보다 뒤일 수 있습니다. 재현 가능한 동작을 원하면 정확한 버전 (예: `@ps-neko/nekowork@0.1.0-alpha.11`) 을 핀하세요.
66
+
67
+ Compatibility / legacy 명령 (`cockpit`, `start`, `ask`, `plan`, `team`, `work`, `verify`, `gate`, `ship`, `run`, `build`, `auto`, `pr-prep`, `report --session`, `apply --session`, `review`) 은 [docs/ADVANCED.md](docs/ADVANCED.md) 에 있습니다. 2.0 에서 제거 예정 ([docs/SCOPE-1.0.md](docs/SCOPE-1.0.md) Phased Cut).
67
68
 
68
69
  ## 한 명령. 하나의 차단된 위험.
69
70
 
71
+ AI 가 작성한 변경에 `process.env.X || "fallback"` 이 들어가면:
72
+
70
73
  ```bash
71
- npx -y @ps-neko/nekowork@alpha auto "add OPENAI_API_KEY fallback for Codex auth"
74
+ npx -y @ps-neko/nekowork@alpha verify-pr
72
75
  ```
73
76
 
74
- 예시 출력:
77
+ 전형적 BLOCK 출력:
75
78
 
76
79
  ```text
77
- Risk: provider-auth / long-lived-secret
78
- Codex verdict: request_changes
79
- Human Gate: required
80
- Ship ready: false
81
- Applied: false
82
-
83
- Blocked because NEKOWORK defaults to delegated CLI auth and rejects long-lived provider API key paths unless the human explicitly opts in.
80
+ === verify-pr ===
81
+ verdict : BLOCK
82
+ reason : Hardcoded secret fallback detected (src/auth.ts:42)
83
+ merge_allowed : false
84
+ apply_allowed : false
85
+ risk_level : CRITICAL
84
86
  ```
85
87
 
86
- 설명: NEKOWORKdelegated CLI auth를 기본값으로 두고, 장기 provider API key 경로는 사람이 명시적으로 선택하지 않는 거부합니다.
87
-
88
- 이것이 NEKOWORK의 핵심입니다. 오토파일럿은 경계 전까지 계속 일할 수 있지만, 위험한 ship/apply 결정은 evidence와 사람의 승인 아래에 둡니다.
88
+ NEKOWORK 의 핵심: AI 변경을 만들 있지만, 위험한 ship/apply 결정은 결정적 룰과 사람 승인 아래에 둡니다. LLM verdict 는 게이트를 통과할 수 없습니다.
89
89
 
90
90
  ## 왜 필요한가
91
91
 
@@ -189,10 +189,10 @@ NEKOWORK는 하나의 거대한 agent 묶음이 아니라, 일을 나누고 검
189
189
  ## 현재 alpha 상태
190
190
 
191
191
  - Package: `@ps-neko/nekowork`
192
- - Current alpha: `0.1.0-alpha.10` candidate
192
+ - Current alpha: `0.1.0-alpha.11` (npm `@alpha` published 2026-05-16)
193
193
  - CLI: `nekowork`
194
194
  - Legacy/internal alias: `harness`
195
- - Tests: 359 pass
195
+ - Tests: 501 pass
196
196
  - npm audit: 0 moderate+ issues
197
197
  - Fresh `npx @alpha` smoke: pass
198
198
 
package/README.md CHANGED
@@ -2,15 +2,23 @@
2
2
 
3
3
  [English](README.md) | [한국어](README.ko.md)
4
4
 
5
- Verifies AI-made code changes before you apply them.
5
+ **Don't merge AI code without verification.**
6
6
 
7
7
  [![validate](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml/badge.svg)](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml)
8
8
 
9
- Bring your coding agent. NEKOWORK proves the change before apply.
9
+ <p align="center">
10
+ <a href="https://ps-neko.github.io/NEKOWORK/?fixture=sample-pr-001">
11
+ <img src="docs/assets/hero.gif" alt="NEKOWORK Verification Factory — Claude advisor 가 LGTM 한 코드를 NEKOWORK 결정적 규칙이 BLOCK 하는 12-station 시연" width="800" />
12
+ </a>
13
+ <br/>
14
+ <em>Claude said LGTM. NEKOWORK blocked.</em> &nbsp;·&nbsp; <a href="https://ps-neko.github.io/NEKOWORK/?fixture=sample-pr-001"><strong>Live demo →</strong></a>
15
+ </p>
10
16
 
11
- NEKOWORK is a local safety gate for AI coding tools. It checks the diff, records evidence, requires a Human Gate for risky work, and only applies a verified change when you explicitly ask it to.
17
+ NEKOWORK is a local verification gate for AI-generated code. It analyzes the diff, runs deterministic risk rules, collects evidence, and decides whether the change is safe to merge or apply without auto-committing, auto-pushing, or trusting LLM verdicts.
12
18
 
13
- Note: "Verified" means independently reviewed with recorded evidence, not mathematically proven correctness. NEKOWORK combines Codex review, test evidence, risk policy, Human Gate, and explicit apply boundaries.
19
+ Note: "Verified" means independently reviewed with recorded evidence not mathematically proven correct. The verdict is decided by deterministic rules and check results. Optional Codex review is recorded as an advisor note only and never controls the verdict.
20
+
21
+ > 1.0 scope and roadmap: [docs/SCOPE-1.0.md](docs/SCOPE-1.0.md). Long-term vision (Verification-first AI development factory): [docs/VISION.md](docs/VISION.md).
14
22
 
15
23
  Note: "ship" in NEKOWORK is a **readiness decision** (`SHIP_READY` or `NO_SHIP`), not a deployment. The `ship` step decides whether `apply` is allowed; it never commits, pushes, deploys, or publishes by itself.
16
24
 
@@ -18,79 +26,85 @@ Default path:
18
26
 
19
27
  ```bash
20
28
  npx -y @ps-neko/nekowork@alpha check
21
- npx -y @ps-neko/nekowork@alpha start "fix failing tests safely" --session first-start
22
- npx -y @ps-neko/nekowork@alpha report --session latest
29
+ npx -y @ps-neko/nekowork@alpha verify-pr
30
+ cat REPORT.md
31
+ cat .nekowork/decision.json
23
32
  ```
24
33
 
25
- Every real `start` run puts the decision first:
34
+ Every real `verify-pr` run puts the verdict first:
26
35
 
27
36
  ```text
28
- Verdict: BLOCKED
29
- Reason: preverify requires Human Gate for secret env fallback
30
- Human Gate: required
31
- Ship ready: false
32
- Apply allowed: false
37
+ === verify-pr ===
38
+ verdict : BLOCK
39
+ reason : Hardcoded secret fallback detected (src/auth.ts:42)
40
+ merge_allowed : false
41
+ apply_allowed : false
42
+ risk_level : CRITICAL
33
43
  ```
34
44
 
35
- The machine-readable companion `decision.json` is shown in [Example Report](#example-report).
45
+ The machine-readable companion `decision.json` and the full report are in [Example Report](#example-report).
36
46
 
37
47
  The evidence chain is intentionally narrow:
38
48
 
39
49
  ```text
40
- diff -> deterministic risk scan -> Codex verification -> decision.json -> REPORT.md -> Human Gate -> explicit apply
50
+ diff -> deterministic risk rules -> available checks (detected, executed in a later alpha) -> evidence package -> deterministic decision -> REPORT.md -> Human Gate -> explicit apply
41
51
  ```
42
52
 
43
- No auto-commit. No auto-push. No surprise deploy. `apply` is explicit and requires verified ship-ready evidence.
53
+ No auto-commit. No auto-push. No surprise deploy. `apply` is explicit; it requires a `decision.json` whose `apply_allowed` is `true`.
44
54
 
45
- Use `start` first. It is the safe beginner entrypoint and prints the final decision before detailed build output. Advanced controls are documented later.
55
+ Bring your AI tool (Cursor / Claude Code / Codex). NEKOWORK starts after the diff is on disk. Advanced and legacy commands are documented in [docs/ADVANCED.md](docs/ADVANCED.md) and gated under Phased Cut (see [docs/SCOPE-1.0.md](docs/SCOPE-1.0.md)).
46
56
 
47
- **Public alpha evidence:** 359 tests / 0 moderate+ npm audit issues / fresh `npx @alpha` smoke / 10 case-study flows / 5 starter packs · [CI badge](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml) · [npm package](https://www.npmjs.com/package/@ps-neko/nekowork) · [terminal transcript](docs/DEMO.md#one-minute-terminal-transcript) · [full report example](docs/DEMO-REPORT.md) · [external run kit](docs/EXTERNAL-RUN.md) · [alpha feedback](https://github.com/Ps-Neko/NEKOWORK/issues/new?template=alpha-feedback.yml) · [roadmap](docs/ROADMAP.md)
57
+ **Public alpha evidence:** 401 tests / 0 moderate+ npm audit issues / fresh `npx @alpha` smoke / 10 case-study flows / 5 starter packs · [CI badge](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml) · [npm package](https://www.npmjs.com/package/@ps-neko/nekowork) · [terminal transcript](docs/DEMO.md#one-minute-terminal-transcript) · [full report example](docs/DEMO-REPORT.md) · [external run kit](docs/EXTERNAL-RUN.md) · [alpha feedback](https://github.com/Ps-Neko/NEKOWORK/issues/new?template=alpha-feedback.yml) · [roadmap](docs/ROADMAP.md)
48
58
 
49
59
  ![NEKOWORK one-minute terminal demo](docs/assets/demo-terminal.svg)
50
60
 
51
61
  ## One Command. One Blocked Risk.
52
62
 
63
+ After your AI tool (Cursor / Claude Code / Codex) writes a `process.env.X || "fallback"` into your auth code, run:
64
+
53
65
  ```bash
54
- npx -y @ps-neko/nekowork@alpha start "add OPENAI_API_KEY fallback for Codex auth"
66
+ npx -y @ps-neko/nekowork@alpha verify-pr
55
67
  ```
56
68
 
57
- Typical blocked-risk evidence:
69
+ Typical blocked-risk output:
58
70
 
59
71
  ```text
60
- Verdict: BLOCKED
61
- Reason: preverify requires Human Gate for secret env fallback
62
- Human Gate: required
63
- Ship ready: false
64
- Apply allowed: false
65
-
66
- Blocked because NEKOWORK defaults to delegated CLI auth and rejects long-lived provider API key paths unless the human explicitly opts in.
72
+ === verify-pr ===
73
+ verdict : BLOCK
74
+ reason : Hardcoded secret fallback detected (src/auth.ts:42)
75
+ risk_level : CRITICAL
76
+ merge_allowed : false
77
+ apply_allowed : false
78
+ findings : critical=1 high=0 medium=0 low=0
79
+ top findings:
80
+ - [CRITICAL] Hardcoded secret fallback detected (src/auth.ts:42)
67
81
  ```
68
82
 
69
- That is the thesis: the coding agent can produce the change, but risky ship/apply decisions stay evidence-backed and human-controlled.
83
+ That is the thesis: AI can write the change, but `verify-pr` runs deterministic rules over the diff and refuses to let unverified changes merge or apply.
70
84
 
71
85
  ## 30-Second First Run
72
86
 
73
- Requirements: Node.js 22+, npm, and git.
87
+ Requirements: Node.js 22+, npm, and git. A git repo with at least one commit.
74
88
 
75
89
  ```bash
76
90
  npx -y @ps-neko/nekowork@alpha check
77
- npx -y @ps-neko/nekowork@alpha start "fix failing tests safely" --session first-start
78
- npx -y @ps-neko/nekowork@alpha report --session latest
91
+ npx -y @ps-neko/nekowork@alpha verify-pr
92
+ cat REPORT.md
93
+ cat .nekowork/decision.json
79
94
  ```
80
95
 
81
- Start with `start` when you want the simplest safe entrypoint. It is the only command a new user needs before reading the report.
96
+ `check` confirms the environment is ready. `verify-pr` scans the current working tree diff with deterministic risk rules, writes evidence to `.nekowork/evidence/`, and decides whether the change is safe to merge or apply. It writes `REPORT.md` at the project root and `.nekowork/decision.json`.
82
97
 
83
98
  Source checkout for local development:
84
99
 
85
100
  ```bash
86
101
  node scripts/cli.js check
87
- node scripts/cli.js start "implement this safely" --session first-start
88
- node scripts/cli.js report --session latest
102
+ node scripts/cli.js verify-pr
89
103
  ```
90
104
 
91
- > **Reproducibility note:** `npx @ps-neko/nekowork@alpha` runs the **published** `0.1.0-alpha.9`. The repository on `main` is the `0.1.0-alpha.10` candidate. If a feature described below was added between those tags, use the source-checkout path above until alpha.10 is published.
105
+ > **Reproducibility note:** `npx @ps-neko/nekowork@alpha` resolves to the most recently published alpha. The published alpha may lag behind `main`. Pin an exact version (e.g. `@ps-neko/nekowork@0.1.0-alpha.11`) for reproducible behavior.
92
106
 
93
- The simple path maps to the evidence loop: `check = doctor --quick`, `start = build`, `report = readable evidence`, and `apply = explicit verified diff application`. See [docs/QUICKSTART.md](docs/QUICKSTART.md) for the longer first-run guide.
107
+ Compatibility / legacy commands (`cockpit`, `start`, `ask`, `plan`, `team`, `work`, `verify`, `gate`, `ship`, `run`, `build`, `auto`, `pr-prep`, `report --session`, `apply --session`, `review`) are documented in [docs/ADVANCED.md](docs/ADVANCED.md). They are scheduled for deprecation in 2.0 per [SCOPE-1.0.md](docs/SCOPE-1.0.md).
94
108
 
95
109
  ## Works With Your Existing AI Workflow
96
110
 
@@ -139,28 +153,35 @@ See the full report contract and example artifact in [docs/DEMO-REPORT.md](docs/
139
153
 
140
154
  ## Main Surface
141
155
 
142
- The user-facing CLI is intentionally small. Three layers:
143
-
144
- **Beginner — start here:**
156
+ **1.0 front surface start here:**
145
157
 
146
158
  - `check` — local readiness probe
147
- - `start` — safe beginner entrypoint, prints verdict first
148
- - `report`readable evidence into `REPORT.md`
149
- - `apply` — explicit verified diff application (refuses without `SHIP_READY` and clear gate)
159
+ - `verify-pr` — verify a diff / PR against deterministic risk rules; writes `REPORT.md` and `.nekowork/decision.json`
160
+ - `verify-pr --comment-file <path>` emit GitHub PR comment markdown for CI integration
161
+ - `verify-pr --ci-exit-soft` — treat `NEEDS_HUMAN_REVIEW` / `INSUFFICIENT_EVIDENCE` as exit 0 (label-driven CI)
162
+
163
+ The CI exit code matrix is fixed:
164
+
165
+ ```text
166
+ ALLOW = 0
167
+ ALLOW_WITH_WARNINGS = 0
168
+ NEEDS_HUMAN_REVIEW = 1
169
+ INSUFFICIENT_EVIDENCE = 1
170
+ BLOCK = 2
171
+ ```
150
172
 
151
- **Advanced phase control:**
173
+ GitHub Actions example: [docs/examples/github-actions-verify-pr.yml](docs/examples/github-actions-verify-pr.yml).
152
174
 
153
- - `ask` / `plan` / `team` / `work` decomposed authoring with single-executor writes
154
- - `verify` / `gate` / `ship` — Codex verification, Human Gate, ship-readiness handoff
155
- - `build` / `auto` / `run` — wrappers over the safety gate; `auto` and `build` never accept `--apply`
156
- - `pr-prep` — review-ready local artifacts without branch, commit, push, or PR
175
+ **Compatibility / labsscheduled for deprecation in 2.0:**
157
176
 
158
- **Legacy compatibility:**
177
+ - Session-based gate: `start` / `report --session` / `apply --session` / `gate status` / `ship --session`
178
+ - Decomposed authoring: `ask` / `plan` / `team` / `work` / `verify` / `pr-prep`
179
+ - Wrappers: `build` / `auto` / `run`
180
+ - Legacy alias: `review` / `review-cycle` / `harness` binary
159
181
 
160
- - `review` / `review-cycle` older full Claude-led / Codex-reviewed workflow
161
- - `harness` binary — legacy alias for `nekowork`
182
+ These commands are functional in alpha and documented in [docs/ADVANCED.md](docs/ADVANCED.md). They will get `[deprecated]` labels in 0.3.x and be removed in 2.0 per [docs/SCOPE-1.0.md](docs/SCOPE-1.0.md). Pure 1.0 users should not need them.
162
183
 
163
- Full stage contract: [docs/CLI-STAGES.md](docs/CLI-STAGES.md). Build modes and routing: [docs/BUILD.md](docs/BUILD.md). Bounded autonomy and the apply boundary: [docs/AUTONOMY.md](docs/AUTONOMY.md). Advanced runtime (`ralph`, `wait`, instincts, cost tracking, Rust supervisor): [docs/ADVANCED.md](docs/ADVANCED.md).
184
+ Stage contract for legacy commands: [docs/CLI-STAGES.md](docs/CLI-STAGES.md). Build modes: [docs/BUILD.md](docs/BUILD.md). Bounded autonomy: [docs/AUTONOMY.md](docs/AUTONOMY.md). Advanced runtime (`ralph`, `wait`, instincts, cost tracking, Rust supervisor): [docs/ADVANCED.md](docs/ADVANCED.md).
164
185
 
165
186
  ## Starter Packs
166
187
 
@@ -186,9 +207,9 @@ For comparison and positioning: [docs/WHY-NEKOWORK.md](docs/WHY-NEKOWORK.md).
186
207
 
187
208
  ## Status
188
209
 
189
- Current repository version: `0.1.0-alpha.10` alpha candidate · Current npm alpha: `@ps-neko/nekowork@0.1.0-alpha.9` (alpha.10 is repository candidate; npm `@alpha` remains 0.1.0-alpha.9 until publish). Package: `@ps-neko/nekowork`. CLI: `nekowork` (`harness` is a legacy alias). Default: mock providers, no API keys.
210
+ Current repository version: `0.1.0-alpha.12` · Current npm alpha: `@ps-neko/nekowork@0.1.0-alpha.11` (published 2026-05-16, `@alpha` dist-tag). Package: `@ps-neko/nekowork`. CLI: `nekowork` (`harness` is a legacy alias). Default: mock providers, no API keys.
190
211
 
191
- Verification: `npm run lint` pass · `npm test` 359 pass · `npm audit --audit-level=moderate` 0 vulns · `npm pack --dry-run --json` pass · `npx -y @ps-neko/nekowork@alpha check` pass with warnings only.
212
+ Verification: `npm run lint` pass · `npm test` 501 tests pass · `npm audit --audit-level=moderate` 0 vulns · `npm pack --dry-run --json` pass · `npx -y @ps-neko/nekowork@alpha check` pass with warnings only.
192
213
 
193
214
  Live provider auth delegates to local CLI sessions (`claude auth status`, `codex login`, `gemini`); long-lived API key env vars (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `GEMINI_API_KEY`, `GOOGLE_API_KEY`) are blocked unless `HARNESS_AUTH_ALLOW_ENV_OVERRIDE=1`. See [docs/SETUP.md](docs/SETUP.md).
194
215
 
@@ -196,7 +217,8 @@ Live provider auth delegates to local CLI sessions (`claude auth status`, `codex
196
217
 
197
218
  - **Core:** [QUICKSTART](docs/QUICKSTART.md) · [CLI-STAGES](docs/CLI-STAGES.md) · [INTEGRATION](docs/INTEGRATION.md) · [UPSTREAM-RECIPES](docs/UPSTREAM-RECIPES.md) · [BUILD](docs/BUILD.md) · [AUTONOMY](docs/AUTONOMY.md) · [SAFETY-GUARANTEES](docs/SAFETY-GUARANTEES.md) · [FAILURE-MODES](docs/FAILURE-MODES.md)
198
219
  - **Demos & evidence:** [DEMO](docs/DEMO.md) · [DEMO-REPORT](docs/DEMO-REPORT.md) · [EXTERNAL-RUN](docs/EXTERNAL-RUN.md) · [case-studies](docs/case-studies)
199
- - **Reference:** [ADVANCED](docs/ADVANCED.md) · [CATALOG-PACKS](docs/CATALOG-PACKS.md) · [PORTING](docs/PORTING.md) · [PR-PREP](docs/PR-PREP.md) · [RELEASE-READINESS](docs/RELEASE-READINESS.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [PRODUCT-PRINCIPLES](docs/PRODUCT-PRINCIPLES.md) · [ROADMAP](docs/ROADMAP.md)
220
+ - **1.0 direction:** [SCOPE-1.0.md](docs/SCOPE-1.0.md) scope, risk rules, decision policy, fixture sourcing · [VISION.md](docs/VISION.md) long-term verification-first OS vision
221
+ - **Reference:** [GUIDED-MODE](docs/GUIDED-MODE.md) · [ADVANCED](docs/ADVANCED.md) · [CATALOG-PACKS](docs/CATALOG-PACKS.md) · [PORTING](docs/PORTING.md) · [PR-PREP](docs/PR-PREP.md) · [RELEASE-READINESS](docs/RELEASE-READINESS.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [PRODUCT-PRINCIPLES](docs/PRODUCT-PRINCIPLES.md) · [ROADMAP](docs/ROADMAP.md)
200
222
  - **Project rules:** [SOUL.md](SOUL.md) · [RULES.md](RULES.md) · [AGENTS.md](AGENTS.md)
201
223
 
202
224
  ## License
@@ -10,7 +10,7 @@
10
10
  ## Current Truth
11
11
 
12
12
  - 위치: `C:/Users/Mun/harness/` · 브랜치: `main`
13
- - 버전: `0.1.0-alpha.10` (repo candidate, npm alpha = 0.1.0-alpha.9)
13
+ - 버전: `0.1.0-alpha.12` (repo; npm alpha `0.1.0-alpha.11` published 2026-05-16, alpha.12 publish 진행 중)
14
14
  - 카탈로그: 11 agents · 5 skills (+1 ralph) · 5 hooks · 6 modules · 5 profiles
15
15
  - 5 빌더 모두 동작 (claude / codex / cursor / gemini / opencode) + codemaps
16
16
  - `npm test`, `npm run lint`, `npm audit --audit-level=moderate`, provider live smoke, Rust release build 검증 경로 유지
@@ -26,14 +26,21 @@
26
26
  ## Active Queues
27
27
 
28
28
  ### In Progress
29
- - 2.5시간 잡티 제거 배치 (CHANGELOG / WORKING-CONTEXT / Validator 경고 / RUNBOOK·PORTING / Security Bar)
29
+ - 외부 알파 5 모집 + 7일 피드백 수집 (POST-RELEASE-CHECKLIST §4-§5, **사용자 수동 social work**)
30
30
 
31
31
  ### Next
32
- - AUDIT 잔존: npm publish 결정, 사내 PoC 결합, OIDC/dead-man/supply-chain 심화
32
+ - 1.0 게이트 5조건 점검 (SCOPE-1.0 §13.2): recall 0.90, FP ≤ 0.10 (real-world corpus), 외부 알파 3/5 "다시 쓰겠다", CRITICAL 미탐 0, 치명적 오탐 0
33
+ - real-world fixture 추가 후 `npm run bench:rules` 재측정
34
+ - 코드 품질 핫스팟 3건 (cli.js 분해 / orchestrator 보일러플레이트 추출 / 미사용 export) — **publish 게이트 통과 이후**
35
+
36
+ ### 절대 금지 (현 단계)
37
+ - 코드 추가 (외부 피드백 없이 추측으로 룰 늘리기 금지)
38
+ - scope 확장 (verify-skill / verify-release 등은 1.x)
39
+ - "1.0 곧 출시" 류 마케팅
33
40
 
34
41
  ## Open PR Classification
35
42
 
36
- (없음레포 push)
43
+ (이전 작업 auth migration / harness.dev placeholder / 검증 게이트 cut 등은 main 머지 완료, 메모리 `nekowork-*` 참조)
37
44
 
38
45
  ## Interfaces
39
46
 
@@ -50,3 +57,6 @@
50
57
  - 2026-04-29: P1 회수 세션 완료 (`docs/dev-log/2026-04-29-p1-recovery.md`). 빈 디렉터리 6 → 0, 미구현 스크립트 9 → 0, ARCHITECTURE 528줄, 73 테스트.
51
58
  - 2026-04-29: 잡티 제거 배치 진행 중 — 본 파일 갱신 + Validator 경고 정합 + RUNBOOK/PORTING/Security Bar 보완.
52
59
  - 2026-04-30: **auth migration 완료**. PR #1-#3 (3계층 인증 + GitHub OAuth + OS keychain) main 머지 (`60e9de9` → `7c4f2c8`, +4 commits, rebase merge). PR #2/#3 은 phase-1 옛 SHA 포함으로 force-push 1회씩(`--onto origin/main bf72841`/`b2b1bce` + `--force-with-lease`). Smoke 3/4 PASS (#1 `claude /status` Claude Max, #2 override 차단 3 케이스, #4 keychain Windows Credential Manager). #3 GitHub OAuth Device Flow 는 OAuth App 미등록으로 사용자 자율 보류 — 실제 GitHub automation 사용 시점에 수행. PR #4 (codex 0.125+ 호환) 는 본 작업과 무관 OPEN 잔존.
60
+ - 2026-05-14: **alpha.10 npm publish 완료**. dist-tag `@alpha`. 19개 CLI 명령 wide surface. 코드 품질 핫스팟 3건 진단 (cli.js 1543 LOC / orchestrator 보일러플레이트 / 미사용 export 3건) — publish 게이트 후 처리.
61
+ - 2026-05-15~16: **1.0 검증 게이트 cut + alpha.11 publish**. `feat(verify-pr)` 5 deterministic rules + Auto-Apply-Commit-Push + GitHub Actions PR comment + bench:rules. 정체성을 "verification-first AI development factory" 12-Station 으로 정제 (VISION.md). README hero 는 검증 게이트 카피 유지.
62
+ - 2026-05-16: **post-publish CI red 사고 복구** (`6a0e862`). `.gitignore *.pem` 룰이 secret-detection 룰 자체 fixture 까지 차단 → alpha.11 가 CI red 6 commit streak 상태로 publish. `!tests/fixtures/**/*.pem` 예외 + synthetic fixture 2개 commit 으로 복구. POST-RELEASE-CHECKLIST §0.2 에 CI green 3항목 게이트 추가 (`f6995ab`). 메모리 `nekowork-alpha11-verify-pr` 의 '자기모순' 섹션 참조.
package/agent.yaml CHANGED
@@ -1,7 +1,7 @@
1
1
  spec_version: gitagent/0.1.0
2
2
  name: nekowork
3
3
  runtime_name: harness
4
- version: 0.1.0-alpha.10
4
+ version: 0.1.0-alpha.12
5
5
  description: "NEKOWORK - Verified autopilot for AI code changes with Codex verification, Human Gate, and explicit apply"
6
6
  license: MIT
7
7
  homepage: https://github.com/Ps-Neko/NEKOWORK
@@ -23,7 +23,7 @@ agents:
23
23
  - doc-writer
24
24
 
25
25
  skills:
26
- - claude-led-codex-review
26
+ - nekowork-full-cycle
27
27
  - plan-eng-review
28
28
  - tdd-workflow
29
29
  - acceptance-coverage
@@ -33,10 +33,11 @@ skills:
33
33
  - release-readiness
34
34
  - porting
35
35
  - ralph # 명시 옵트인 영속 루프. 자동 키워드 활성 OFF.
36
+ - verified-gate # strict 검증 게이트 진입점. 명시 옵트인, 자동 키워드 OFF.
36
37
 
37
38
  commands:
38
39
  # legacy compat (slash entry). 신규 추가 금지.
39
- - claude-led-codex-review
40
+ - nekowork-full-cycle
40
41
 
41
42
  hooks:
42
43
  file: hooks/hooks.json
@@ -200,7 +201,7 @@ auth:
200
201
  deny_static_api_keys_in_repo: true
201
202
 
202
203
  routing:
203
- # 단계별 routing 표는 skills/claude-led-codex-review/SKILL.md 가 정전(canon).
204
+ # 단계별 routing 표는 skills/nekowork-full-cycle/SKILL.md 가 정전(canon).
204
205
  eco_mode_floor: sonnet
205
206
  human_gate_triggers:
206
207
  severity: critical
@@ -5,7 +5,7 @@ provider: codex
5
5
  model: gpt-5-codex
6
6
  level: 3
7
7
  disallowedTools: [Write, Edit, Bash, Network]
8
- trigger: ["codex challenge", "--secure", "claude-led-codex-review:6"]
8
+ trigger: ["codex challenge", "--secure", "nekowork-full-cycle:6"]
9
9
  hand_off_to: []
10
10
  sandbox: read-only
11
11
  network_access: false
@@ -5,7 +5,7 @@ provider: codex
5
5
  model: gpt-5-codex
6
6
  level: 3
7
7
  disallowedTools: [Write, Edit, Bash, Network]
8
- trigger: ["codex review", "claude-led-codex-review:5"]
8
+ trigger: ["codex review", "nekowork-full-cycle:5"]
9
9
  hand_off_to: []
10
10
  sandbox: read-only
11
11
  network_access: false
@@ -0,0 +1,29 @@
1
+ ---
2
+ description: "Claude 주도 + Codex 위임 7단계 풀사이클. nekowork-full-cycle 스킬 호출."
3
+ ---
4
+
5
+ # /nekowork-full-cycle
6
+
7
+ 이 슬래시 명령은 `nekowork-full-cycle` 스킬의 legacy compat 진입점이다. 신규 워크플로우는 스킬에서 정의되지만 슬래시 호출 호환성을 위해 보존.
8
+
9
+ ## 동작
10
+
11
+ `Skill` 도구로 `nekowork-full-cycle` 를 즉시 호출. 인자가 있으면 작업 요약으로 전달, 없으면 사용자에게 한 줄 요약을 요청.
12
+
13
+ ## 인자
14
+
15
+ - `$ARGUMENTS` — 작업 요약 한 줄
16
+ - `--fast` — 단계 1·6 스킵
17
+ - `--secure` — 단계 6 강제
18
+ - `--no-ship` — 단계 7 생략
19
+
20
+ ## 예시
21
+
22
+ ```
23
+ /nekowork-full-cycle JWT 검증 미들웨어 추가 --secure
24
+ /nekowork-full-cycle 결제 환불 로직 버그 수정
25
+ /nekowork-full-cycle --fast 사소한 리팩토링
26
+ /nekowork-full-cycle 새 API 엔드포인트 --no-ship
27
+ ```
28
+
29
+ 전체 명세는 `skills/nekowork-full-cycle/SKILL.md` 참조.
package/docs/ADVANCED.md CHANGED
@@ -1,5 +1,21 @@
1
1
  # Advanced Features
2
2
 
3
+ > ## Phased Cut status
4
+ >
5
+ > 이 페이지의 대부분 명령은 알파.10 시기의 wide CLI surface 입니다.
6
+ > NEKOWORK 1.0 은 [검증 게이트 정체성](SCOPE-1.0.md) 에 집중하며,
7
+ > 이 명령들은 [Phased Cut](SCOPE-1.0.md#2-phased-cut-단계) 을 거칩니다.
8
+ >
9
+ > | Phase | 시기 | 상태 |
10
+ > |---|---|---|
11
+ > | Phase 0 | now (0.1.x → 0.2.x) | **functional** + hero 강등 |
12
+ > | Phase 1 | 0.3.x → 1.0 | `[deprecated]` 마크 + "removed in 2.0" 경고 |
13
+ > | Phase 2 | 1.x → 2.0 | 제거 또는 `@ps-neko/nekowork-legacy` 분리 |
14
+ >
15
+ > 1.0 의 hero 명령 4종 — `check / verify-pr / report / apply` — 으로 이주를 권장합니다.
16
+ > 명령별 운명은 [SCOPE-1.0.md §3](SCOPE-1.0.md#3-명령-운명-표) 참고.
17
+ > 장기 비전은 [VISION.md](VISION.md) 참고.
18
+
3
19
  The public alpha path focuses on `doctor`, `build`, `report`, `gate`, and the decomposed `ask`, `plan`, `team`, `work`, `verify`, `ship`, `apply`, `run`, `review`, `review-cycle`, and install/apply surfaces. This page keeps the larger runtime surface discoverable without crowding the first-run docs.
4
20
 
5
21
  ## team