@ps-neko/nekowork 0.1.0-alpha.7 → 0.1.0-alpha.8
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.
- package/CLAUDE.md +3 -3
- package/README.md +133 -93
- package/agent.yaml +3 -2
- package/docs/ADVANCED.md +2 -1
- package/docs/ARCHITECTURE.md +4 -3
- package/docs/AUDIT.md +11 -11
- package/docs/AUTONOMY.md +92 -0
- package/docs/BUILD.md +23 -10
- package/docs/CATALOG-PACKS.md +25 -16
- package/docs/CHANGELOG.md +20 -1
- package/docs/CLI-STAGES.md +11 -2
- package/docs/CODEMAPS/manifests.md +1 -0
- package/docs/CODEMAPS/schemas.md +1 -0
- package/docs/CODEMAPS/scripts.md +10 -0
- package/docs/CODEMAPS/tests.md +2 -0
- package/docs/DEMO-REPORT.md +14 -0
- package/docs/DEMO.md +1 -1
- package/docs/PARALLEL-CANDIDATES.md +58 -0
- package/docs/PORTING.md +1 -1
- package/docs/PR-PREP.md +35 -0
- package/docs/PRODUCT-PRINCIPLES.md +11 -2
- package/docs/PUBLISH-ALPHA.md +28 -14
- package/docs/QUICKSTART.md +20 -2
- package/docs/RELEASE-READINESS.md +34 -20
- package/docs/ROADMAP.md +52 -1
- package/docs/SETUP.md +1 -1
- package/docs/WHY-NEKOWORK.md +7 -6
- package/docs/assets/demo-terminal.svg +1 -1
- package/manifests/build-modes.json +61 -0
- package/manifests/install-profiles.json +61 -0
- package/package.json +4 -2
- package/schemas/build-modes.schema.json +42 -0
- package/scripts/ci/validate-manifests.js +26 -1
- package/scripts/cli/commands/auto-command.js +198 -0
- package/scripts/cli/commands/build-command.js +259 -0
- package/scripts/cli.js +23 -240
- package/scripts/lib/build-modes.js +38 -0
- package/scripts/orchestrators/auto.js +263 -0
- package/scripts/orchestrators/build.js +31 -8
- package/scripts/orchestrators/report.js +49 -6
package/CLAUDE.md
CHANGED
|
@@ -8,7 +8,7 @@
|
|
|
8
8
|
|
|
9
9
|
## 자동 갱신 영역
|
|
10
10
|
|
|
11
|
-
<!-- HARNESS:START version=0.1.0-alpha.
|
|
11
|
+
<!-- HARNESS:START version=0.1.0-alpha.8 -->
|
|
12
12
|
<!-- 이 영역은 scripts/sync-claude-md.js 가 자동 갱신한다. 직접 편집 금지. -->
|
|
13
13
|
|
|
14
14
|
## 카탈로그 요약
|
|
@@ -17,8 +17,8 @@
|
|
|
17
17
|
- skills: 10
|
|
18
18
|
- commands: 1 (legacy compat)
|
|
19
19
|
- hooks: 5 (gateguard-fact-force, config-protection, quality-gate, pre-bash-dispatcher, persistent-mode)
|
|
20
|
-
- packs: core, builder, quality, security, frontend, testing, release, enterprise
|
|
21
|
-
- profiles: core, developer, builder, security, product, quality, frontend, testing, research, full
|
|
20
|
+
- packs: core, builder, productivity, team, debugging, maintenance, pr, catalog-plus, quality, security, frontend, testing, release, enterprise
|
|
21
|
+
- profiles: core, developer, builder, productivity, security, product, quality, frontend, testing, research, full
|
|
22
22
|
- harnesses: claude, codex, cursor, gemini, opencode
|
|
23
23
|
|
|
24
24
|
## 에이전트 → 모델 매트릭스
|
package/README.md
CHANGED
|
@@ -1,27 +1,36 @@
|
|
|
1
1
|
# NEKOWORK
|
|
2
2
|
|
|
3
|
-
|
|
3
|
+
Verified Autopilot for AI code changes.
|
|
4
4
|
|
|
5
5
|
[](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml)
|
|
6
6
|
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
7
|
+
AI builds. Codex verifies. You approve the boundary.
|
|
8
|
+
|
|
9
|
+
NEKOWORK plans, edits, verifies, repairs, and prepares ship-ready AI code changes. Final apply remains human-controlled.
|
|
10
|
+
|
|
11
|
+
It runs:
|
|
12
|
+
|
|
13
|
+
1. Autonomous planning and build
|
|
14
|
+
2. Independent Codex verification
|
|
15
|
+
3. Bounded repair when findings are fixable
|
|
16
|
+
4. Report, ship/no-ship, and Human Gate
|
|
17
|
+
5. Explicit apply only when the human chooses it
|
|
18
|
+
|
|
19
|
+
No auto-commit. No auto-push. No surprise deploy.
|
|
17
20
|
|
|
18
21
|
Product principle:
|
|
19
22
|
|
|
20
23
|
```text
|
|
21
|
-
NEKOWORK =
|
|
22
|
-
```
|
|
24
|
+
NEKOWORK = verified autopilot -> Codex verification -> Human Gate -> explicit apply
|
|
25
|
+
```
|
|
26
|
+
|
|
27
|
+
```text
|
|
28
|
+
Autonomous until apply.
|
|
29
|
+
Verified before ship.
|
|
30
|
+
Human-controlled at the boundary.
|
|
31
|
+
```
|
|
23
32
|
|
|
24
|
-
NEKOWORK packages a local runtime with one source catalog, `agent.yaml`, projected into Claude Code, Codex CLI, Cursor, Gemini CLI, and OpenCode surfaces. The `harness` CLI remains a legacy/internal alias for `nekowork`.
|
|
33
|
+
NEKOWORK packages a local runtime with one source catalog, `agent.yaml`, projected into Claude Code, Codex CLI, Cursor, Gemini CLI, and OpenCode surfaces. The `harness` CLI remains a legacy/internal alias for `nekowork`.
|
|
25
34
|
|
|
26
35
|
NEKOWORK is intentionally not a 100-agent pack. Every agent, skill, hook, profile, module, and pack must:
|
|
27
36
|
|
|
@@ -30,9 +39,13 @@ NEKOWORK is intentionally not a 100-agent pack. Every agent, skill, hook, profil
|
|
|
30
39
|
3. produce auditable evidence,
|
|
31
40
|
4. respect Human Gate.
|
|
32
41
|
|
|
33
|
-
**Public alpha evidence:**
|
|
42
|
+
**Public alpha evidence:** 14 packs / 11 profiles / 36 components / 5 harness targets / 7 case-study flows / 290 tests / 0 moderate+ npm audit issues / fresh `npx @alpha` smoke
|
|
34
43
|
|
|
35
|
-
NEKOWORK does not automatically commit, push, publish, deploy, or apply diffs. `apply` is explicit and requires verified ship-ready evidence.
|
|
44
|
+
NEKOWORK does not automatically commit, push, publish, deploy, or apply diffs. `apply` is explicit and requires verified ship-ready evidence.
|
|
45
|
+
|
|
46
|
+
For bounded autonomy before that boundary, use `auto`: it can route, build, verify, repair fixable findings within a budget, write a report, and then stop before apply.
|
|
47
|
+
|
|
48
|
+
Next track: `auto --parallel-candidates N` will let isolated candidate workers propose patches, then NEKOWORK will compare them into one canonical ship candidate before Codex verification and Human Gate.
|
|
36
49
|
|
|
37
50
|
**Latest alpha evidence:** [CI badge](https://github.com/Ps-Neko/NEKOWORK/actions/workflows/harness-validate.yml) / [npm package](https://www.npmjs.com/package/@ps-neko/nekowork) / [smoke transcript](docs/DEMO.md#one-minute-terminal-transcript) / [report artifact](docs/DEMO-REPORT.md)
|
|
38
51
|
|
|
@@ -40,27 +53,33 @@ NEKOWORK does not automatically commit, push, publish, deploy, or apply diffs. `
|
|
|
40
53
|
|
|
41
54
|

|
|
42
55
|
|
|
43
|
-
##
|
|
44
|
-
|
|
45
|
-
Use the current npm alpha for the
|
|
46
|
-
|
|
47
|
-
```bash
|
|
48
|
-
npx -y @ps-neko/nekowork@alpha check
|
|
49
|
-
npx -y @ps-neko/nekowork@alpha
|
|
50
|
-
npx -y @ps-neko/nekowork@alpha
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
56
|
+
## 30-Second First Run
|
|
57
|
+
|
|
58
|
+
Use the current npm alpha for the fastest proof of the workflow:
|
|
59
|
+
|
|
60
|
+
```bash
|
|
61
|
+
npx -y @ps-neko/nekowork@alpha check
|
|
62
|
+
npx -y @ps-neko/nekowork@alpha auto "fix failing tests safely" --session first-auto
|
|
63
|
+
npx -y @ps-neko/nekowork@alpha report --session latest
|
|
64
|
+
```
|
|
65
|
+
|
|
66
|
+
Start with `auto` when you want NEKOWORK to keep going until report/gate. Use `build` when you want one build pass. Drop down to `work`, `verify`, and `ship` only when you need phase-level control.
|
|
67
|
+
|
|
68
|
+
Preview the route before running providers or writing session state:
|
|
69
|
+
|
|
70
|
+
```bash
|
|
71
|
+
npx -y @ps-neko/nekowork@alpha auto "fix failing tests safely" --dry-run
|
|
72
|
+
npx -y @ps-neko/nekowork@alpha build "fix this safely" --dry-run
|
|
73
|
+
```
|
|
55
74
|
|
|
56
75
|
Use a source checkout for local development:
|
|
57
76
|
|
|
58
77
|
```bash
|
|
59
|
-
node scripts/cli.js check
|
|
60
|
-
node scripts/cli.js
|
|
61
|
-
node scripts/cli.js report --session latest
|
|
62
|
-
node scripts/cli.js gate status --session latest
|
|
63
|
-
```
|
|
78
|
+
node scripts/cli.js check
|
|
79
|
+
node scripts/cli.js auto "implement this safely" --session first-auto
|
|
80
|
+
node scripts/cli.js report --session latest
|
|
81
|
+
node scripts/cli.js gate status --session latest
|
|
82
|
+
```
|
|
64
83
|
|
|
65
84
|
Or use the decomposed beginner path directly:
|
|
66
85
|
|
|
@@ -71,7 +90,7 @@ node scripts/cli.js report --session first-run
|
|
|
71
90
|
node scripts/cli.js gate status --session first-run
|
|
72
91
|
```
|
|
73
92
|
|
|
74
|
-
The simple paths map to the evidence loop: `check = doctor --quick`, `build = auto routing plus mode presets over run`, and `run = work -> verify -> ship`.
|
|
93
|
+
The simple paths map to the evidence loop: `check = doctor --quick`, `build = auto routing plus mode presets over run`, `auto = bounded build/verify/repair/report before apply`, and `run = work -> verify -> ship`.
|
|
75
94
|
|
|
76
95
|
Use `build --dry-run` when you want to preview auto routing, mode, profile, workers, stages, and apply policy before running providers or writing session state. Use `build --explain` when you want the same routing rationale and evidence list after a real build.
|
|
77
96
|
|
|
@@ -96,14 +115,16 @@ Strict quality: enabled
|
|
|
96
115
|
Acceptance coverage: 4/5
|
|
97
116
|
Quality warnings: 2
|
|
98
117
|
|
|
99
|
-
Evidence:
|
|
100
|
-
- work-summary.json
|
|
101
|
-
- verify-summary.json
|
|
102
|
-
- ship-summary.json
|
|
103
|
-
- gate-summary.json
|
|
104
|
-
```
|
|
105
|
-
|
|
106
|
-
|
|
118
|
+
Evidence:
|
|
119
|
+
- work-summary.json
|
|
120
|
+
- verify-summary.json
|
|
121
|
+
- ship-summary.json
|
|
122
|
+
- gate-summary.json
|
|
123
|
+
```
|
|
124
|
+
|
|
125
|
+
The first screen of `REPORT.md` is the trust card: work produced, independent verification, Human Gate, ship readiness, apply state, and whether the target project was mutated.
|
|
126
|
+
|
|
127
|
+
See the full report contract and example artifact in [docs/DEMO-REPORT.md](docs/DEMO-REPORT.md), and the one-minute terminal transcript in [docs/DEMO.md](docs/DEMO.md).
|
|
107
128
|
|
|
108
129
|
## Human Gate Example
|
|
109
130
|
|
|
@@ -147,29 +168,31 @@ Apply command: node scripts/cli.js apply --session first-work
|
|
|
147
168
|
|---|---|---|
|
|
148
169
|
| Large Claude Code packs | More agents, commands, skills | Curated verification loop |
|
|
149
170
|
| Team simulation | More specialist perspectives | Read-only team plus one executor |
|
|
150
|
-
| Autopilot | Fast autonomous execution |
|
|
171
|
+
| Autopilot | Fast autonomous execution | verified autonomy until apply, report, gate, explicit apply |
|
|
151
172
|
| Discipline workflows | Better development habits | Evidence-backed ship decision |
|
|
152
173
|
|
|
153
|
-
## When To Choose
|
|
154
|
-
|
|
155
|
-
| Use case |
|
|
156
|
-
|---|---|
|
|
157
|
-
|
|
|
158
|
-
|
|
|
159
|
-
|
|
|
160
|
-
|
|
|
161
|
-
|
|
|
162
|
-
|
|
|
163
|
-
|
|
164
|
-
|
|
174
|
+
## When To Choose NEKOWORK
|
|
175
|
+
|
|
176
|
+
| Use case | NEKOWORK fit |
|
|
177
|
+
|---|---|
|
|
178
|
+
| You want one command to keep working until report/gate | `auto` routes, builds, verifies, repairs, and stops before apply |
|
|
179
|
+
| You want one build pass with safe routing | `build` routes the task into safe mode presets |
|
|
180
|
+
| You want daily planning, TDD, debugging, and finish checks | use the `productivity` pack |
|
|
181
|
+
| You want team-style review before implementation | use the `team` pack; handoffs stay read-only |
|
|
182
|
+
| You need PR or release evidence | use `pr` or `release` before ship/apply |
|
|
183
|
+
| You need sensitive-change control | use `security` and keep Human Gate active |
|
|
184
|
+
| You need explicit apply instead of autopilot mutation | keep the default `report -> gate -> apply` path |
|
|
185
|
+
|
|
186
|
+
Use other AI development tools when they fit your preferred authoring flow. Use NEKOWORK when AI work needs to become verified, reportable, gated, and explicitly applied.
|
|
165
187
|
|
|
166
188
|
## Three Paths
|
|
167
189
|
|
|
168
190
|
Most users should start with the Beginner path. The other paths are for explicit phase control or legacy compatibility.
|
|
169
191
|
|
|
170
|
-
1. Beginner: `check ->
|
|
171
|
-
2.
|
|
172
|
-
3.
|
|
192
|
+
1. Beginner verified autopilot: `check -> auto -> report -> gate`
|
|
193
|
+
2. One-pass safe build: `check -> build -> report -> gate`
|
|
194
|
+
3. Advanced: `ask -> plan -> team -> work -> verify -> gate -> ship -> report -> apply`
|
|
195
|
+
4. Legacy: `review` / `review-cycle`
|
|
173
196
|
|
|
174
197
|
## Why NEKOWORK
|
|
175
198
|
|
|
@@ -177,19 +200,19 @@ NEKOWORK is for teams that want AI-assisted development without making the agent
|
|
|
177
200
|
|
|
178
201
|
## Status
|
|
179
202
|
|
|
180
|
-
- Current repository version: `0.1.0-alpha.
|
|
181
|
-
- Current package name: `@ps-neko/nekowork`
|
|
182
|
-
- Published CLI names: `nekowork` and `harness`
|
|
183
|
-
- Current npm alpha: `@ps-neko/nekowork@0.1.0-alpha.
|
|
184
|
-
- Current npm alpha.
|
|
185
|
-
- Supported install path today: npm alpha, clone, submodule, or local repository integration
|
|
203
|
+
- Current repository version: `0.1.0-alpha.8` alpha candidate
|
|
204
|
+
- Current package name: `@ps-neko/nekowork`
|
|
205
|
+
- Published CLI names: `nekowork` and `harness`
|
|
206
|
+
- Current npm alpha: `@ps-neko/nekowork@0.1.0-alpha.7`
|
|
207
|
+
- Current npm alpha.8 status: repository candidate; public publish is pending owner OTP/web auth
|
|
208
|
+
- Supported install path today: npm alpha, clone, submodule, or local repository integration
|
|
186
209
|
- Dist-tag note: use `@alpha` until a stable release; `latest` still points at the first alpha line
|
|
187
210
|
- Default mode: mock providers, no API keys, no provider CLI calls
|
|
188
211
|
|
|
189
212
|
Current local verification:
|
|
190
213
|
|
|
191
214
|
- `npm run lint`: pass
|
|
192
|
-
- `npm test`:
|
|
215
|
+
- `npm test`: 290 tests pass
|
|
193
216
|
- `npm audit --audit-level=moderate`: 0 vulnerabilities
|
|
194
217
|
- `npm pack --dry-run --json`: pass
|
|
195
218
|
- `npx -y @ps-neko/nekowork@alpha check`: pass with warnings only
|
|
@@ -208,12 +231,19 @@ Current local verification:
|
|
|
208
231
|
|
|
209
232
|
## Official Packs
|
|
210
233
|
|
|
211
|
-
| Pack | Adds | Use when |
|
|
212
|
-
|---|---|---|
|
|
213
|
-
| `core` | minimal verification runtime | first install or repo smoke |
|
|
214
|
-
| `
|
|
215
|
-
| `
|
|
216
|
-
| `
|
|
234
|
+
| Pack | Adds | Use when |
|
|
235
|
+
|---|---|---|
|
|
236
|
+
| `core` | minimal verification runtime | first install or repo smoke |
|
|
237
|
+
| `builder` | safe build modes entrypoint | one-command build with verification and gates |
|
|
238
|
+
| `productivity` | planning, TDD, debugging, finish routines | daily AI-assisted development |
|
|
239
|
+
| `team` | read-only role handoffs | you want team-style review before one executor writes |
|
|
240
|
+
| `debugging` | failing-test and regression triage | the task starts from a bug or unclear root cause |
|
|
241
|
+
| `maintenance` | dependency, refactor, migration, cleanup routines | routine upkeep still needs verification |
|
|
242
|
+
| `pr` | diff review, test evidence, changelog, risk notes | preparing or reviewing a PR |
|
|
243
|
+
| `catalog-plus` | richest curated catalog surface | evaluating the full NEKOWORK catalog |
|
|
244
|
+
| `quality` | acceptance coverage, strict evidence prompts | feature work needs proof |
|
|
245
|
+
| `security` | auth/secrets/deploy risk prompts | sensitive changes |
|
|
246
|
+
| `frontend` | UI mockup, component review, accessibility checks | product-facing UI work |
|
|
217
247
|
| `testing` | regression planning and coverage handoffs | test confidence is the main risk |
|
|
218
248
|
| `release` | ship/no-ship evidence | pre-release checks |
|
|
219
249
|
| `enterprise` | full catalog with all gates | high-control teams |
|
|
@@ -316,9 +346,10 @@ The public alpha surface is intentionally small:
|
|
|
316
346
|
- `gate`: inspect, approve, or block a human gate for a session
|
|
317
347
|
- `ship`: produce a ship/no-ship readiness handoff after Codex verification
|
|
318
348
|
- `apply`: apply a verified `SHIP_READY` live-work diff to the target project
|
|
319
|
-
- `run`: execute the decomposed wrapper, `work -> verify -> ship`, with optional apply
|
|
320
|
-
- `build`: one-command builder wrapper with default `auto` routing, explicit `fast`, `safe`, `team`, `tdd`, `release`, and `--dry-run` preview
|
|
321
|
-
- `
|
|
349
|
+
- `run`: execute the decomposed wrapper, `work -> verify -> ship`, with optional apply
|
|
350
|
+
- `build`: one-command builder wrapper with default `auto` routing, explicit `fast`, `safe`, `team`, `tdd`, `release`, and `--dry-run` preview
|
|
351
|
+
- `auto`: bounded autonomy wrapper that can repair fixable no-ship findings within budget, then report and stop before apply
|
|
352
|
+
- `report`: summarize session evidence into `REPORT.md` without project mutation
|
|
322
353
|
- `review`: run the legacy full Claude-led/Codex-reviewed workflow
|
|
323
354
|
- `review-cycle`: explicit compatibility alias for the legacy full review workflow
|
|
324
355
|
- `install --plan` / `install --apply`: project generated harness surfaces
|
|
@@ -327,22 +358,27 @@ Advanced features such as `team-lite`, `ralph`, `wait`, instincts, cost tracking
|
|
|
327
358
|
|
|
328
359
|
`plan` is recommended before `work` for larger changes. The current `run` command intentionally stays compact: it runs `work -> verify -> ship`, records acceptance criteria through `work`, and applies only when `--apply` is explicitly provided.
|
|
329
360
|
|
|
330
|
-
Use `build "<task>"` when NEKOWORK should be the single entrypoint. It defaults to `--mode auto`, classifies the task, selects `fast`, `safe`, `team`, `tdd`, or `release`, records build intelligence, and still uses one executor for writes, Codex verification before ship, and explicit apply only. Use an explicit `--mode` when you need to override the router.
|
|
331
|
-
|
|
332
|
-
Risky explicit overrides are protected. For example, `build "change OAuth token validation" --mode fast` is blocked because auto routing recommends `safe
|
|
361
|
+
Use `build "<task>"` when NEKOWORK should be the single entrypoint. It defaults to `--mode auto`, classifies the task, selects `fast`, `safe`, `team`, `tdd`, or `release`, records build intelligence, and still uses one executor for writes, Codex verification before ship, and explicit apply only. The mode safety ordering is manifest-backed in `manifests/build-modes.json`. Use an explicit `--mode` when you need to override the router.
|
|
362
|
+
|
|
363
|
+
Risky explicit overrides are protected. For example, `build "change OAuth token validation" --mode fast` is blocked because auto routing recommends `safe`, and `build "prepare npm package publish release notes" --mode fast` is blocked because auto routing recommends the higher-safety `release` mode. Use the recommended mode or add `--force-mode` only when you intentionally accept that downgrade.
|
|
364
|
+
|
|
365
|
+
Use `auto "<task>"` when NEKOWORK should continue before the apply boundary. `auto` routes through the same build intelligence, runs `build`, repeats fixable no-ship work within `--level cautious|normal|aggressive` budgets, writes `auto-summary.json`, generates `REPORT.md`, and never accepts `--apply`.
|
|
333
366
|
|
|
334
367
|
Use `--profile quality` or `--profile security` on `work`, `verify`, and `run` when a task needs stronger evidence prompts. Add `--strict-quality` to `verify`, `run`, or `build` when missing evidence or acceptance coverage should become a fix-required verdict before ship.
|
|
335
368
|
|
|
336
369
|
Use official packs when choosing an install shape:
|
|
337
370
|
|
|
338
371
|
```bash
|
|
339
|
-
node scripts/install-plan.js --list
|
|
340
|
-
node scripts/install-plan.js --pack
|
|
341
|
-
node scripts/install-plan.js --pack
|
|
342
|
-
node scripts/install-plan.js --pack
|
|
343
|
-
|
|
344
|
-
|
|
345
|
-
|
|
372
|
+
node scripts/install-plan.js --list
|
|
373
|
+
node scripts/install-plan.js --pack productivity
|
|
374
|
+
node scripts/install-plan.js --pack team
|
|
375
|
+
node scripts/install-plan.js --pack pr
|
|
376
|
+
node scripts/install-plan.js --pack builder
|
|
377
|
+
node scripts/install-plan.js --pack quality
|
|
378
|
+
node scripts/install-plan.js --pack security --target codex --json
|
|
379
|
+
```
|
|
380
|
+
|
|
381
|
+
Packs are aliases over validated profiles. They add clearer product packaging without weakening the core gates. `productivity` is the shortest daily discipline pack: brainstorm, plan, TDD, debug, execute, verify, report, and finish over the same safe build loop. `team`, `debugging`, `maintenance`, `pr`, and `catalog-plus` make the catalog feel richer while still resolving to safety-checked profiles.
|
|
346
382
|
|
|
347
383
|
## Catalog
|
|
348
384
|
|
|
@@ -350,8 +386,8 @@ Packs are aliases over validated profiles. They add clearer product packaging wi
|
|
|
350
386
|
- Skills: 10
|
|
351
387
|
- Hooks: 5
|
|
352
388
|
- Modules: 7
|
|
353
|
-
- Profiles: `core`, `developer`, `builder`, `security`, `product`, `quality`, `frontend`, `testing`, `research`, `full`
|
|
354
|
-
- Official packs: `core`, `builder`, `quality`, `security`, `frontend`, `testing`, `release`, `enterprise`
|
|
389
|
+
- Profiles: `core`, `developer`, `builder`, `productivity`, `security`, `product`, `quality`, `frontend`, `testing`, `research`, `full`
|
|
390
|
+
- Official packs: `core`, `builder`, `productivity`, `team`, `debugging`, `maintenance`, `pr`, `catalog-plus`, `quality`, `security`, `frontend`, `testing`, `release`, `enterprise`
|
|
355
391
|
- Harness targets: `claude`, `codex`, `cursor`, `gemini`, `opencode`
|
|
356
392
|
|
|
357
393
|
Key skills:
|
|
@@ -372,9 +408,10 @@ Key skills:
|
|
|
372
408
|
```bash
|
|
373
409
|
node scripts/cli.js doctor
|
|
374
410
|
node scripts/cli.js doctor --quick --gemini-smoke
|
|
375
|
-
npm run demo:quick
|
|
376
|
-
node scripts/cli.js build "builder smoke" --mode team --session build-smoke
|
|
377
|
-
node scripts/cli.js
|
|
411
|
+
npm run demo:quick
|
|
412
|
+
node scripts/cli.js build "builder smoke" --mode team --session build-smoke
|
|
413
|
+
node scripts/cli.js auto "fix failing tests safely" --level normal --dry-run
|
|
414
|
+
node scripts/cli.js report --session latest
|
|
378
415
|
node scripts/install-plan.js --list
|
|
379
416
|
node scripts/install-plan.js --pack quality
|
|
380
417
|
node scripts/install-plan.js --profile developer
|
|
@@ -419,13 +456,16 @@ npm run security:hardening
|
|
|
419
456
|
npm pack --dry-run --json
|
|
420
457
|
```
|
|
421
458
|
|
|
422
|
-
`npm pack --dry-run --json` currently produces a package named like `ps-neko-nekowork-0.1.0-alpha.
|
|
459
|
+
`npm pack --dry-run --json` currently produces a package named like `ps-neko-nekowork-0.1.0-alpha.8.tgz`. It does not publish.
|
|
423
460
|
|
|
424
461
|
## Documentation
|
|
425
462
|
|
|
426
|
-
- [docs/QUICKSTART.md](docs/QUICKSTART.md) - first run and common paths
|
|
427
|
-
- [docs/BUILD.md](docs/BUILD.md) - build command modes and invariants
|
|
428
|
-
- [docs/
|
|
463
|
+
- [docs/QUICKSTART.md](docs/QUICKSTART.md) - first run and common paths
|
|
464
|
+
- [docs/BUILD.md](docs/BUILD.md) - build command modes and invariants
|
|
465
|
+
- [docs/AUTONOMY.md](docs/AUTONOMY.md) - bounded autonomy, repair budgets, and the apply boundary
|
|
466
|
+
- [docs/PARALLEL-CANDIDATES.md](docs/PARALLEL-CANDIDATES.md) - planned isolated candidate writer contract
|
|
467
|
+
- [docs/PR-PREP.md](docs/PR-PREP.md) - planned PR prep artifact contract
|
|
468
|
+
- [docs/WHY-NEKOWORK.md](docs/WHY-NEKOWORK.md) - comparison and product positioning
|
|
429
469
|
- [docs/CATALOG-PACKS.md](docs/CATALOG-PACKS.md) - curated catalog, official packs, and case-study evidence
|
|
430
470
|
- [docs/PUBLISH-ALPHA.md](docs/PUBLISH-ALPHA.md) - public npm alpha release plan
|
|
431
471
|
- [docs/ROADMAP.md](docs/ROADMAP.md) - small alpha roadmap and non-goals
|
package/agent.yaml
CHANGED
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
spec_version: gitagent/0.1.0
|
|
2
2
|
name: nekowork
|
|
3
3
|
runtime_name: harness
|
|
4
|
-
version: 0.1.0-alpha.
|
|
5
|
-
description: "NEKOWORK -
|
|
4
|
+
version: 0.1.0-alpha.8
|
|
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
|
|
8
8
|
# authors: contributor 목록은 git 히스토리로 갈음. 별도 명시 안 함.
|
|
@@ -90,6 +90,7 @@ profiles:
|
|
|
90
90
|
- core
|
|
91
91
|
- developer
|
|
92
92
|
- builder
|
|
93
|
+
- productivity
|
|
93
94
|
- security
|
|
94
95
|
- product
|
|
95
96
|
- quality
|
package/docs/ADVANCED.md
CHANGED
|
@@ -205,7 +205,8 @@ Rules:
|
|
|
205
205
|
- `release` focuses on ship/readiness and report evidence.
|
|
206
206
|
- `--dry-run` previews auto routing, preset resolution, stages, workers, and safety invariants without writing session state.
|
|
207
207
|
- `--explain` prints routing rationale and evidence files after the build.
|
|
208
|
-
- `--force-mode` is required when a risky task is manually forced
|
|
208
|
+
- `--force-mode` is required when a risky task is manually forced into a lower-safety mode than the risk-aware recommendation.
|
|
209
|
+
- Build mode safety ranks are defined in `manifests/build-modes.json` and validated with the manifest schemas, so mode policy changes are reviewable outside the CLI code.
|
|
209
210
|
- `apply` is never implicit; use `--apply` only for verified live-work diffs.
|
|
210
211
|
|
|
211
212
|
Outputs:
|
package/docs/ARCHITECTURE.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# Architecture
|
|
2
2
|
|
|
3
|
-
NEKOWORK is the product and public name. It
|
|
3
|
+
NEKOWORK is the product and public name. It is a verified autopilot for AI code changes with one canonical catalog projected into multiple agent surfaces. The `harness` binary remains a legacy/internal CLI alias.
|
|
4
4
|
|
|
5
5
|
## Core Idea
|
|
6
6
|
|
|
@@ -88,6 +88,7 @@ node scripts/cli.js report --session work-smoke --project-root <target>
|
|
|
88
88
|
node scripts/cli.js apply --session work-smoke --project-root <target>
|
|
89
89
|
node scripts/cli.js run "decomposed wrapper" --session run-smoke --project-root <target>
|
|
90
90
|
node scripts/cli.js build "safe builder wrapper" --mode team --session build-smoke --project-root <target>
|
|
91
|
+
node scripts/cli.js auto "bounded autonomy before apply" --level normal --session auto-smoke --project-root <target>
|
|
91
92
|
node scripts/cli.js review "change request" --no-ship --project-root <target>
|
|
92
93
|
node scripts/cli.js review-cycle "legacy full-cycle request" --no-ship --project-root <target>
|
|
93
94
|
```
|
|
@@ -205,8 +206,8 @@ Builders project the catalog into tool-specific files:
|
|
|
205
206
|
|
|
206
207
|
## Release State
|
|
207
208
|
|
|
208
|
-
The current repository release line is `0.1.0-alpha.
|
|
209
|
+
The current repository release line is `0.1.0-alpha.8` candidate:
|
|
209
210
|
|
|
210
211
|
- Repository and GitHub tarball release are available.
|
|
211
|
-
- Public npm alpha is published as `@ps-neko/nekowork@alpha` and currently points at `0.1.0-alpha.
|
|
212
|
+
- Public npm alpha is published as `@ps-neko/nekowork@alpha` and currently points at `0.1.0-alpha.7`; alpha.8 publish is pending owner OTP/web auth.
|
|
212
213
|
- Clone, submodule, and local checkout integration remain supported for repository-pinned workflows.
|
package/docs/AUDIT.md
CHANGED
|
@@ -2,20 +2,20 @@
|
|
|
2
2
|
|
|
3
3
|
Status date: 2026-05-08
|
|
4
4
|
|
|
5
|
-
This audit summarizes the current NEKOWORK state after
|
|
5
|
+
This audit summarizes the current NEKOWORK state after preparing the `0.1.0-alpha.8` repository alpha candidate. Public npm `@alpha` remains on `0.1.0-alpha.7` until the owner completes OTP/web auth for alpha.8.
|
|
6
6
|
|
|
7
7
|
## Current Status
|
|
8
8
|
|
|
9
9
|
| Area | Status | Notes |
|
|
10
10
|
|---|---|---|
|
|
11
|
-
| Package metadata | OK | repository version `@ps-neko/nekowork@0.1.0-alpha.
|
|
12
|
-
| npm publish |
|
|
11
|
+
| Package metadata | OK | repository version `@ps-neko/nekowork@0.1.0-alpha.8`, `agent.yaml` uses `name: nekowork`, `runtime_name: harness`, matching version, and `nekowork`/`harness` CLI bins |
|
|
12
|
+
| npm publish | Pending | `@ps-neko/nekowork@alpha` points at `0.1.0-alpha.7`; alpha.8 dry-run passes but real publish is blocked by owner OTP |
|
|
13
13
|
| Source install | OK | Clone, local checkout, and submodule workflows are documented |
|
|
14
|
-
| Public npm alpha | OK | `docs/PUBLISH-ALPHA.md` records alpha publishes through `0.1.0-alpha.
|
|
14
|
+
| Public npm alpha | OK | `docs/PUBLISH-ALPHA.md` records alpha publishes through `0.1.0-alpha.7` and alpha.8 candidate state |
|
|
15
15
|
| CLI doctor/check | OK | `check`, `doctor`, `doctor --quick`, and `doctor --gemini-smoke` are available |
|
|
16
16
|
| Provider auth | OK | Local delegated CLI auth is the default path |
|
|
17
17
|
| Internal provider adapter | OK | `HARNESS_PROVIDER_OVERRIDE=internal` can call an explicit JSON command adapter without weakening gates |
|
|
18
|
-
| Catalog | OK |
|
|
18
|
+
| Catalog | OK | 14 official packs, 11 agents, 10 skills, 5 hooks, 7 modules, 36 components, 11 profiles |
|
|
19
19
|
| Multi-harness output | OK | Claude, Codex, Cursor, Gemini, and OpenCode builders are present |
|
|
20
20
|
| Quick demo | OK | `npm run demo:quick` verifies the shortest no-API `doctor -> build -> report -> gate status` path |
|
|
21
21
|
| Fresh npm alpha smoke | OK | CI runs `npx -y @ps-neko/nekowork@alpha check --json` from a disposable directory |
|
|
@@ -30,7 +30,7 @@ This audit summarizes the current NEKOWORK state after publishing the `0.1.0-alp
|
|
|
30
30
|
| Persistent wakeup | OK | `wait` resumes supported active sessions and blocks on `HUMAN_GATE` |
|
|
31
31
|
| Generated docs | OK | CODEMAP output is stable ASCII and reproducible |
|
|
32
32
|
| Tests | OK | Unit, integration, and e2e suites pass locally and in CI |
|
|
33
|
-
| Release |
|
|
33
|
+
| Release | Pending | `v0.1.0-alpha.7` is the latest GitHub prerelease; alpha.8 tag/release waits on publish |
|
|
34
34
|
|
|
35
35
|
## Verification Gates
|
|
36
36
|
|
|
@@ -55,7 +55,7 @@ Current local result for this working tree:
|
|
|
55
55
|
- `npm run test:unit`: covered by full `npm test`
|
|
56
56
|
- `npm run validate:all`: pass
|
|
57
57
|
- `npm run lint`: pass
|
|
58
|
-
- `npm test`:
|
|
58
|
+
- `npm test`: 290 tests pass
|
|
59
59
|
- quick run demo: pass through `npm run demo:quick -- --cleanup`
|
|
60
60
|
- external project e2e smoke: pass through `npm test`
|
|
61
61
|
- `node scripts/sync-claude-md.js --check`: pass
|
|
@@ -63,9 +63,9 @@ Current local result for this working tree:
|
|
|
63
63
|
- `npm audit --audit-level=moderate`: 0 vulnerabilities
|
|
64
64
|
- `npm pack --dry-run --json`: pass
|
|
65
65
|
- `npm publish --dry-run --access public --tag alpha`: pass
|
|
66
|
-
- `npm publish --access public --tag alpha`: `
|
|
67
|
-
- `npm view @ps-neko/nekowork dist-tags version versions --json`: `alpha` points at `0.1.0-alpha.
|
|
68
|
-
- `npx -y @ps-neko/nekowork@alpha check`: passed for `0.1.0-alpha.
|
|
66
|
+
- `npm publish --access public --tag alpha`: blocked by `EOTP`; owner OTP/web auth still required
|
|
67
|
+
- `npm view @ps-neko/nekowork dist-tags version versions --json`: `alpha` points at `0.1.0-alpha.7`; `latest` remains `0.1.0-alpha.0`
|
|
68
|
+
- `npx -y @ps-neko/nekowork@alpha check`: passed for `0.1.0-alpha.7` with WARN summary from Gemini auth not checked
|
|
69
69
|
|
|
70
70
|
## Completed Work
|
|
71
71
|
|
|
@@ -94,7 +94,7 @@ Current local result for this working tree:
|
|
|
94
94
|
- Official packs expose curated install shapes without creating a second safety model.
|
|
95
95
|
- Checked-in example fixtures now cover financial UI, CI hardening, and quality lifecycle evidence flows.
|
|
96
96
|
- Third-party case studies record NEKOWORK runs against `sindresorhus/is-plain-obj`, `jshttp/basic-auth`, `python-hyper/h11`, and `motdotla/dotenv`.
|
|
97
|
-
- Public npm alpha `0.1.0-alpha.
|
|
97
|
+
- Public npm alpha `0.1.0-alpha.7` is published under the `alpha` dist-tag; alpha.8 is a repository candidate.
|
|
98
98
|
|
|
99
99
|
## Remaining Optional Work
|
|
100
100
|
|
package/docs/AUTONOMY.md
ADDED
|
@@ -0,0 +1,92 @@
|
|
|
1
|
+
# Verified Autopilot
|
|
2
|
+
|
|
3
|
+
NEKOWORK is a verified autopilot for AI code changes. It can plan, build, verify, and repair before the apply boundary, but it never applies, commits, pushes, publishes, or deploys without explicit human action.
|
|
4
|
+
|
|
5
|
+
Autonomy is bounded by the apply boundary:
|
|
6
|
+
|
|
7
|
+
```text
|
|
8
|
+
route -> build -> verify -> repair loop -> report -> Human Gate / explicit apply
|
|
9
|
+
```
|
|
10
|
+
|
|
11
|
+
`auto` can plan, build, verify, and repair fixable findings before apply. It never commits, pushes, publishes, deploys, opens a PR, or applies a diff.
|
|
12
|
+
|
|
13
|
+
## Command
|
|
14
|
+
|
|
15
|
+
```bash
|
|
16
|
+
nekowork auto "fix failing tests safely"
|
|
17
|
+
nekowork auto "implement OAuth login" --level cautious
|
|
18
|
+
nekowork auto "prepare release readiness" --level normal --mode release
|
|
19
|
+
nekowork auto "large cleanup" --level aggressive --budget 5
|
|
20
|
+
```
|
|
21
|
+
|
|
22
|
+
Preview without creating a session:
|
|
23
|
+
|
|
24
|
+
```bash
|
|
25
|
+
nekowork auto "change OAuth token validation" --dry-run --json
|
|
26
|
+
```
|
|
27
|
+
|
|
28
|
+
## Levels
|
|
29
|
+
|
|
30
|
+
| Level | Repair Budget | Use when |
|
|
31
|
+
|---|---:|---|
|
|
32
|
+
| `cautious` | 1 round, no repair loop | the human wants one verified attempt and a report |
|
|
33
|
+
| `normal` | up to 3 rounds | fixable findings can be repaired before report |
|
|
34
|
+
| `aggressive` | up to 5 rounds | larger work can iterate more, but still stops before apply |
|
|
35
|
+
|
|
36
|
+
All levels preserve the same hard boundary:
|
|
37
|
+
|
|
38
|
+
- no automatic `apply`
|
|
39
|
+
- no automatic commit, push, publish, deploy, or PR creation
|
|
40
|
+
- multi-worker thinking stays read-only
|
|
41
|
+
- one executor owns project-file mutation per work round
|
|
42
|
+
- Codex verification remains required before ship/apply
|
|
43
|
+
- Human Gate cannot be bypassed
|
|
44
|
+
|
|
45
|
+
## Repair Policy
|
|
46
|
+
|
|
47
|
+
`auto` repeats the safe build loop only when the prior round is fixable:
|
|
48
|
+
|
|
49
|
+
| Result | Auto behavior |
|
|
50
|
+
|---|---|
|
|
51
|
+
| `ship_ready` | stop and write report |
|
|
52
|
+
| `no_ship` with fixable findings | repair until the level budget is exhausted |
|
|
53
|
+
| `human_gate` | stop immediately |
|
|
54
|
+
| unknown or non-fixable state | stop and write report |
|
|
55
|
+
|
|
56
|
+
The output is `auto-summary.json` plus the normal session evidence:
|
|
57
|
+
|
|
58
|
+
```text
|
|
59
|
+
build-intelligence.json
|
|
60
|
+
build-plan.json
|
|
61
|
+
acceptance-criteria.json
|
|
62
|
+
build-summary.json
|
|
63
|
+
run-summary.json
|
|
64
|
+
verify-summary.json
|
|
65
|
+
ship-summary.json
|
|
66
|
+
REPORT.md
|
|
67
|
+
```
|
|
68
|
+
|
|
69
|
+
## Apply Boundary
|
|
70
|
+
|
|
71
|
+
`auto` deliberately rejects `--apply`.
|
|
72
|
+
|
|
73
|
+
After `auto` finishes, the human should inspect:
|
|
74
|
+
|
|
75
|
+
```bash
|
|
76
|
+
nekowork report --session <id>
|
|
77
|
+
nekowork gate status --session <id>
|
|
78
|
+
```
|
|
79
|
+
|
|
80
|
+
Only after verified `SHIP_READY` evidence and clear gates should the human choose an explicit apply command:
|
|
81
|
+
|
|
82
|
+
```bash
|
|
83
|
+
nekowork apply --session <id>
|
|
84
|
+
```
|
|
85
|
+
|
|
86
|
+
This is the product rule:
|
|
87
|
+
|
|
88
|
+
```text
|
|
89
|
+
Autonomous until apply.
|
|
90
|
+
Verified before ship.
|
|
91
|
+
Human-controlled at the boundary.
|
|
92
|
+
```
|