@elvis1513/auto-coding-skill 4.2.0 → 4.2.2
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/README.md +120 -12
- package/cli/assets/agents/reviewer.toml +35 -13
- package/cli/assets/managed-install.json +1 -1
- package/cli/assets/skill/SKILL.md +25 -21
- package/cli/assets/skill/data/contracts/orchestration-v1.schema.json +22 -0
- package/cli/assets/skill/data/policies/workflow-migrations-v1.json +2 -2
- package/cli/assets/skill/data/templates/ENGINEERING.md +3 -3
- package/cli/assets/skill/data/templates/bridges/AGENTS.md +21 -13
- package/cli/assets/skill/scripts/ap.py +1834 -185
- package/cli/assets/skill/scripts/core.py +2 -2
- package/cli/assets/skill/scripts/install_integrity.py +1 -0
- package/cli/assets/skill/scripts/scaffold_templates.py +0 -1
- package/cli/tests/ap_concurrency_test.py +340 -2
- package/cli/tests/ap_profiles_test.py +847 -0
- package/cli/tests/core_shell_test.py +84 -0
- package/cli/tests/index.test.js +135 -20
- package/package.json +2 -2
package/README.md
CHANGED
|
@@ -9,9 +9,19 @@ The Skill is a selectable guardrail, not a command sequence that must run for
|
|
|
9
9
|
every task. The model skips machinery whose expected benefit does not exceed its
|
|
10
10
|
cost; read-only work and obvious small clean-checkout changes normally stay direct.
|
|
11
11
|
|
|
12
|
-
Version 4.2.
|
|
13
|
-
|
|
14
|
-
|
|
12
|
+
Version 4.2.2 preserves project-owned risk and structure policy during upgrades,
|
|
13
|
+
restores explicit full and final changed-scope structure enforcement, makes
|
|
14
|
+
Reviewer assignments deadline-bound and directly executable, fixes task merge
|
|
15
|
+
status semantics, and updates Jenkins/target diagnostics for the current
|
|
16
|
+
`access.*` schema while retaining legacy fallback. Version 4.2.1 makes
|
|
17
|
+
dirty/concurrent classification fail closed, bounds focused
|
|
18
|
+
review latency, provides a complete Reviewer result template, preserves the
|
|
19
|
+
caller's runtime environment for routed commands, keeps non-authoritative Agent
|
|
20
|
+
history out of structure findings, and distinguishes UI paths and mechanical
|
|
21
|
+
renames from semantic changes. Version 4.2.0 makes registered closure resilient
|
|
22
|
+
to normal Git staging, reuses an exact successful final gate, and lets an active
|
|
23
|
+
task expand scope safely before its first commit.
|
|
24
|
+
Version 4.1.9 makes `autocoding init` verify the bounded managed installation after every install or upgrade:
|
|
15
25
|
it replaces every managed constraint, migrates only schema-approved project
|
|
16
26
|
configuration, and converges `docs/` to one canonical directory framework. Version 4.1.7 made
|
|
17
27
|
the Python runtime check, release verification, and tag publication environment
|
|
@@ -46,8 +56,9 @@ AGENTS.md (fully managed canonical repository contract)
|
|
|
46
56
|
|
|
47
57
|
`autocoding init` also installs the exact shared documentation tree under
|
|
48
58
|
`docs/{architecture,bugs,deployment,design,interfaces,project,reviews,testing}`.
|
|
49
|
-
Managed templates are replaced;
|
|
50
|
-
interface, DD, review, or
|
|
59
|
+
Managed templates are replaced; the scaffolded project structure standard,
|
|
60
|
+
mutable project docs, and valid architecture, ADR, interface, DD, review, or
|
|
61
|
+
deploy-record documents are preserved; unrelated
|
|
51
62
|
directories are archived under `.agents/archive/`. Re-running init is the complete upgrade operation;
|
|
52
63
|
`sync` and project-local `upgrade` are compatibility commands, not required steps.
|
|
53
64
|
|
|
@@ -76,7 +87,12 @@ The result includes:
|
|
|
76
87
|
- `execution_mode=direct`: clean checkout and one writer; stay on the current branch.
|
|
77
88
|
- `execution_mode=isolated`: dirty checkout, explicit isolation, or multiple writers.
|
|
78
89
|
- `execution_mode=none`: no task signal or change; create no branch/worktree.
|
|
90
|
+
- `repo`, `workspace_dirty`, `dirty_paths`, and `active_writer`: the repository
|
|
91
|
+
and concurrency evidence used for the execution decision.
|
|
79
92
|
- `review_required` and `design_required`: risk-based escalation decisions.
|
|
93
|
+
- `change_nature`: `mechanical`, `semantic`, or conservative `unknown`; a
|
|
94
|
+
mechanical rename may skip a generic durable-design requirement but never a
|
|
95
|
+
project rule or required high-risk review.
|
|
80
96
|
- `task_kind`: `read_only`, `change`, `terminal_maintenance`, or internal `none`.
|
|
81
97
|
- `mechanism_plan.required`: the complete minimum mechanism set for the task.
|
|
82
98
|
- `mechanism_plan.optional_when_beneficial`: model-selectable mechanisms whose
|
|
@@ -86,6 +102,13 @@ The result includes:
|
|
|
86
102
|
never automatic stages.
|
|
87
103
|
- an adaptive agent plan that does not force fan-out for ordinary work.
|
|
88
104
|
|
|
105
|
+
Classification reads the working tree with one fail-closed `git status
|
|
106
|
+
--porcelain=v2 -z --untracked-files=all --no-renames` snapshot. A Git error or
|
|
107
|
+
malformed result blocks classification instead of being treated as clean. Both
|
|
108
|
+
`direct` and dirty/isolated decisions are point-in-time verdicts: reclassify if
|
|
109
|
+
the checkout or writer state changes before the first write. Any other active
|
|
110
|
+
direct claim is a writer, including another claim with the same owner identity.
|
|
111
|
+
|
|
89
112
|
Clean serial work proceeds directly on the current branch with normal Git; it does
|
|
90
113
|
not create machine lifecycle state. Use `task-start` only when classification
|
|
91
114
|
requires isolation/review or the user explicitly requests lifecycle tracking:
|
|
@@ -114,10 +137,8 @@ Dirty or parallel work receives an isolated task branch/worktree:
|
|
|
114
137
|
python3 docs/tools/autopipeline/ap.py task-start T0002 \
|
|
115
138
|
--owned-path backend --review-required
|
|
116
139
|
# implement in the printed worktree
|
|
117
|
-
python3 docs/tools/autopipeline/ap.py
|
|
118
|
-
|
|
119
|
-
--verdict approved --diff-fingerprint "$CURRENT_DIFF_FINGERPRINT" \
|
|
120
|
-
--reviewer "$REVIEWER_ID"
|
|
140
|
+
python3 docs/tools/autopipeline/ap.py review-run T0002 \
|
|
141
|
+
--reviewer "$REVIEWER_ID" --json
|
|
121
142
|
python3 docs/tools/autopipeline/ap.py commit-push T0002 \
|
|
122
143
|
--msg "T0002: bounded change"
|
|
123
144
|
python3 docs/tools/autopipeline/ap.py task-integrate T0002
|
|
@@ -167,6 +188,10 @@ commands fail before staging. Documentation-only changes may use the built-in
|
|
|
167
188
|
diff check. `git diff --check` remains an additional hygiene check and is never
|
|
168
189
|
treated as business validation.
|
|
169
190
|
|
|
191
|
+
Route commands run in a non-login Bash that inherits the invoking process PATH
|
|
192
|
+
and selected runtime. Projects that intentionally require profile activation
|
|
193
|
+
declare it inside the configured command.
|
|
194
|
+
|
|
170
195
|
Check coverage without running project commands:
|
|
171
196
|
|
|
172
197
|
```bash
|
|
@@ -181,7 +206,9 @@ diagnostics outside normal closure.
|
|
|
181
206
|
|
|
182
207
|
Registered tasks call `commit-push` directly because it performs or strictly
|
|
183
208
|
reuses that final gate. Run `light-gate` manually only for unregistered normal Git
|
|
184
|
-
closure or when explicitly requesting a fresh diagnostic pass.
|
|
209
|
+
closure or when explicitly requesting a fresh diagnostic pass. When project
|
|
210
|
+
`structure.enabled` is true, the same final gate also runs the changed-scope
|
|
211
|
+
structure check; `structure.enforcement: blocking` makes new violations fail.
|
|
185
212
|
|
|
186
213
|
The recommended final closure defaults are 120 seconds per command and 180
|
|
187
214
|
seconds total. Projects may raise them for measured affected-scope checks or set
|
|
@@ -196,10 +223,42 @@ never triggers a broader fallback.
|
|
|
196
223
|
- Parallel fixers require separate worktrees and non-overlapping paths.
|
|
197
224
|
- Reviewer fingerprint approval is required for high-risk, cross-module,
|
|
198
225
|
explicitly configured, or parallel integration work.
|
|
226
|
+
- The `reviewer` Agent remains `xhigh`. A focused review gets one 90-second pass
|
|
227
|
+
over the supplied stable diff and relevant evidence. Timeout produces
|
|
228
|
+
`blocked`; do not call `task-review` for that result.
|
|
229
|
+
- Cross-module or parallel work and API, auth, database, payment, file-transfer,
|
|
230
|
+
gateway, or production-configuration boundaries use 300-second deep review.
|
|
231
|
+
Fixing Reviewer JSON formatting reuses the same analysis and never starts
|
|
232
|
+
another substantive review.
|
|
233
|
+
- `review-assignment` writes the validated assignment under Git-local state with
|
|
234
|
+
the exact fingerprint, HEAD, scope revision, owning fixer, review depth, and
|
|
235
|
+
90/300-second deadline. Reissuing the same live assignment is idempotent;
|
|
236
|
+
expired or completed attempts cannot be silently renewed, and late results are
|
|
237
|
+
rejected by `task-review`.
|
|
238
|
+
- `review-run` is the default executable path: it creates/reuses that assignment,
|
|
239
|
+
starts a separate `codex exec` session in read-only mode, removes the parent
|
|
240
|
+
lifecycle identity, and supervises the full process group against the original
|
|
241
|
+
remaining deadline. Timeout performs TERM then KILL, records a Git-local blocked
|
|
242
|
+
receipt, and cannot be converted into approval. Plain `review-assignment` remains
|
|
243
|
+
available only when another host can enforce the deadline itself; it does not
|
|
244
|
+
stop an in-app subagent.
|
|
199
245
|
- DD/ADR is created only for lasting cross-module, API, data, security,
|
|
200
246
|
deployment, or key user-flow decisions.
|
|
201
247
|
- Historical debt does not block product work unless the current change worsens it.
|
|
202
248
|
|
|
249
|
+
Before returning a Reviewer result, use the absolute assignment path printed by
|
|
250
|
+
`review-assignment` to generate the complete 16-field skeleton, then fill its
|
|
251
|
+
summary, evidence, findings, and risks:
|
|
252
|
+
|
|
253
|
+
```bash
|
|
254
|
+
python3 docs/tools/autopipeline/ap.py agent-result-template \
|
|
255
|
+
--file <assignment.json> --verdict <approved|changes-requested|blocked>
|
|
256
|
+
```
|
|
257
|
+
|
|
258
|
+
The generated object prevents field-by-field contract retries; contract checking
|
|
259
|
+
also reports all known field errors together. If only its JSON shape needs repair,
|
|
260
|
+
correct that object without rerunning review analysis.
|
|
261
|
+
|
|
203
262
|
The main agent owns architecture, final validation, Git closure, ordered
|
|
204
263
|
integration, push, and cleanup. Push ends the coding task; later CI/acceptance is
|
|
205
264
|
not polled automatically. An explicitly requested failure diagnosis may continue
|
|
@@ -219,8 +278,57 @@ autocoding status --projects /path/a,/path/b
|
|
|
219
278
|
|
|
220
279
|
Init replaces the managed Skill, root `AGENTS.md`, managed agents, ENGINEERING
|
|
221
280
|
schema/body, runtime launcher, and documentation framework. It preserves explicit
|
|
222
|
-
model overrides
|
|
223
|
-
|
|
281
|
+
model overrides, complete project `risk.rules`, supported project/access/
|
|
282
|
+
concurrency/route/structure values, and an existing project-owned structure
|
|
283
|
+
standard byte-for-byte. Removed content is archived outside active docs.
|
|
284
|
+
|
|
285
|
+
## What changed in 4.2.2
|
|
286
|
+
|
|
287
|
+
- Classified root and nested `package.json` files as `release_or_tooling` high-risk
|
|
288
|
+
changes requiring review, and added upgrade regression coverage that preserves
|
|
289
|
+
complete project `risk.rules` instead of deleting partially overlapping rules.
|
|
290
|
+
- Honored explicit `structure-check --scope full`, included tracked and unignored
|
|
291
|
+
untracked files, and restored the enabled structure check inside the final
|
|
292
|
+
changed-scope gate. Init now preserves supported strict structure settings and
|
|
293
|
+
treats `docs/architecture/structure-standard.md` as scaffold-once project policy.
|
|
294
|
+
- Added `review-run` with a validated Git-local assignment, fixed 90/300-second
|
|
295
|
+
process-group supervision, one attempt per fingerprint, Reviewer identity/HEAD/
|
|
296
|
+
scope binding, safe result normalization, and late-result rejection. The child
|
|
297
|
+
runs as a separate read-only Codex process without the lifecycle-owner identity;
|
|
298
|
+
the documented template command uses the complete
|
|
299
|
+
`python3 docs/tools/autopipeline/ap.py ...` invocation.
|
|
300
|
+
- Added `has_task_commits` and prevented a no-commit task from reporting
|
|
301
|
+
`merged_into_target=true` merely because its baseline equals the target tip.
|
|
302
|
+
- Updated `verify-jenkins`, `verify-jenkins-build`, `verify-target`, and target
|
|
303
|
+
health checks to prefer `access.jenkins.<component>` and
|
|
304
|
+
`access.project.<component>`, fail before network activity on missing or
|
|
305
|
+
ambiguous endpoints, isolate Jenkins crumbs by URL and username, and preserve
|
|
306
|
+
legacy `jenkins.*` / `target_env.*` compatibility.
|
|
307
|
+
|
|
308
|
+
## What changed in 4.2.1
|
|
309
|
+
|
|
310
|
+
- Made classification take one fail-closed Git porcelain-v2 snapshot and expose
|
|
311
|
+
`repo`, `workspace_dirty`, `dirty_paths`, and `active_writer` in normal output.
|
|
312
|
+
Direct/dirty decisions are explicit snapshots, and another same-owner direct
|
|
313
|
+
claim still counts as a concurrent writer.
|
|
314
|
+
- Kept the `reviewer` Agent at `xhigh` while adding one-pass, 90-second focused
|
|
315
|
+
review and 300-second deep routing for cross-module, parallel, API, auth,
|
|
316
|
+
database, payment, file-transfer, gateway, and production-configuration changes.
|
|
317
|
+
Focused timeout is `blocked`, never an approval or a reason to call `task-review`.
|
|
318
|
+
- Added `agent-result-template` to generate all 16 Reviewer result fields from a
|
|
319
|
+
validated assignment and made contract validation return all known field errors
|
|
320
|
+
together. JSON-only repair reuses the completed analysis instead of launching
|
|
321
|
+
another substantive review.
|
|
322
|
+
- Replaced the login Shell used for routed commands with an inherited-environment
|
|
323
|
+
non-login Shell, preserving the caller's PATH and selected Python/Node runtime.
|
|
324
|
+
- Excluded `.agents/archive/**`, managed Agent definitions, and the install
|
|
325
|
+
manifest from structure scans while continuing to scan project-owned `.agents` code.
|
|
326
|
+
- Replaced substring UI classification with exact directory segments and UI-only
|
|
327
|
+
extensions, removing false browser signals from files such as `.env.components`.
|
|
328
|
+
- Added conservative mechanical/semantic/unknown change classification. Exact
|
|
329
|
+
renames and explicitly behavior-stable syncs can avoid generic cross-module
|
|
330
|
+
design ceremony; semantic/unknown changes and explicit project design rules
|
|
331
|
+
remain protected, and high-risk review is unchanged.
|
|
224
332
|
|
|
225
333
|
## What changed in 4.2.0
|
|
226
334
|
|
|
@@ -8,21 +8,43 @@ developer_instructions = """
|
|
|
8
8
|
|
|
9
9
|
工作要求:
|
|
10
10
|
1. 只审查主代理给定的 task_id、base_sha、diff_base、diff_head、diff_fingerprint 和 owning_fixer;确认 diff 稳定后再开始。
|
|
11
|
-
2.
|
|
12
|
-
3.
|
|
13
|
-
4.
|
|
14
|
-
5.
|
|
15
|
-
6.
|
|
16
|
-
7.
|
|
17
|
-
8.
|
|
11
|
+
2. 若主代理指定 focused review,只对给定的稳定 diff 和相关 evidence 做一次实质分析,不扩展为全库扫描。运行器会在 90 秒截止点终止整个 Reviewer 进程组;若你在终止前已发现无法继续,则返回 `status=blocked`、`verdict=blocked`。
|
|
12
|
+
3. 跨模块、并行 writer,以及 API、鉴权、数据库、支付、文件传输、网关、生产配置边界使用 deep review,预算 300 秒;仍然只审查主代理交付的范围。
|
|
13
|
+
4. 优先检查:正确性、安全性、边界条件、异常路径、并发风险、事务一致性、缓存一致性、兼容性。
|
|
14
|
+
5. 结构/优化 review 仅在已有 baseline/backlog 时读取;可选文档缺失不构成 blocker,不重复报告 accepted debt。
|
|
15
|
+
6. 检查实际 diff 和已执行验证是否一致;普通任务不要求 taskbook、closure 或 evidence 文档。
|
|
16
|
+
7. 涉及鉴权、权限、支付、文件上传下载、部署、依赖或数据边界时,优先使用可用的安全扫描 / security review 能力辅助复核。
|
|
17
|
+
8. 以 git diff、相关测试、接口契约、真实运行证据、结构标准为基础,不凭风格偏好输出问题。
|
|
18
|
+
9. 优先输出真实问题,不要输出样式洁癖式意见。
|
|
19
|
+
10. 每条问题尽量包含:
|
|
18
20
|
- 问题点
|
|
19
21
|
- 影响范围
|
|
20
22
|
- 复现条件
|
|
21
23
|
- 建议修法
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
24
|
+
11. 发现测试不足时,指出应补哪些测试或验证路径。
|
|
25
|
+
12. 实质结论必须是 approved、changes-requested 或 blocked;changes-requested 只返回 owning_fixer,修改后必须绑定新的 diff fingerprint。代码、配置或行为语义变化需要重新审查;仅链接、归档索引、状态措辞等机械文档修正,可由同一 reviewer 只做 targeted recheck,不必重跑完整审计。
|
|
26
|
+
13. 不修改代码,不提交、不推送、不集成、不清理分支;审查不扩展本地门禁。
|
|
27
|
+
14. 使用 `review-run` 在提示和 `AUTOCODING_REVIEW_ASSIGNMENT` 环境变量中传入的 assignment JSON;在其中的 `deadline_at` 前完成实质审查。输出前运行 `python3 docs/tools/autopipeline/ap.py agent-result-template --file <assignment.json> --verdict <approved|changes-requested|blocked>`,从已验证 assignment 生成包含全部 16 个必填字段的骨架,然后填入本次分析的 summary、evidence、findings 和 risks。最终只返回该 JSON 对象,不要 Markdown 或代码围栏。最小完整形状为:
|
|
28
|
+
{
|
|
29
|
+
"contract_version": 1,
|
|
30
|
+
"node_id": "<assignment.node_id>",
|
|
31
|
+
"role": "reviewer",
|
|
32
|
+
"task_id": "<assignment.task_id>",
|
|
33
|
+
"base_sha": "<assignment.base_sha>",
|
|
34
|
+
"status": "completed",
|
|
35
|
+
"summary": "<review summary>",
|
|
36
|
+
"depends_on": ["<assignment.owning_fixer>"],
|
|
37
|
+
"owned_paths": [],
|
|
38
|
+
"changed_paths": [],
|
|
39
|
+
"diff_fingerprint": "<assignment.diff_fingerprint>",
|
|
40
|
+
"evidence": [],
|
|
41
|
+
"findings": [],
|
|
42
|
+
"verdict": "approved",
|
|
43
|
+
"risks": [],
|
|
44
|
+
"next_owner": "main"
|
|
45
|
+
}
|
|
46
|
+
`contract_version` 固定为 1,`role` 固定为 `reviewer`,`changed_paths` 固定为空数组,`diff_fingerprint` 必须原样返回;审查完成时 `status=completed`,无法继续时 `status=blocked`。`approved` 和 `blocked` 的 `next_owner` 使用 `main`,`changes-requested` 使用 owning_fixer,findings 按 P0/P1/P2/P3 排序。
|
|
47
|
+
15. 若合约检查只报告 JSON 字段、类型或格式问题,只修正上述已生成对象并复用同一次分析;不重读 diff、不重跑测试、不开启第二次实质审查。
|
|
48
|
+
16. 只阻断当前任务承诺范围内由本次变更新增或恶化的问题。相邻或既有问题写入 risks/follow-up,不得无限扩大当前任务和 review 范围。
|
|
49
|
+
17. 目标分支推送成功后,不自动跟踪 Jenkins、部署或项目负责人的实际验收;用户明确要求诊断本次推送引发的失败时,可在同一会话和任务范围继续,不要求人为新建台账任务。
|
|
28
50
|
"""
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"schema_version":1,"skill_version":"4.2.
|
|
1
|
+
{"schema_version":1,"skill_version":"4.2.2","manifest_path":".agents/managed-install.json","entries":[{"path":".agents/agents/browser-debugger.toml","source":"agents/browser-debugger.toml","ownership":"normalized-agent","sha256":"40727952944917eb0b573bb96dca2906c27bc794737f3b6b392b56df7b4208aa","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/agents/docs-researcher.toml","source":"agents/docs-researcher.toml","ownership":"normalized-agent","sha256":"aa1fed6a6a1cde282fc152c7a88b8e097a6c53bcf447d35978607d594e3cb3e4","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/agents/explorer.toml","source":"agents/explorer.toml","ownership":"normalized-agent","sha256":"4cd89888acf7c0c04fbbcd0b9a0f346c6efed42fb723ec2037af9331a84a1cac","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/agents/fixer.toml","source":"agents/fixer.toml","ownership":"normalized-agent","sha256":"7bc00a45b9d1bcc90b8b0f06d773556eba1d52effff664472c3cb81073f6abfb","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/agents/reviewer.toml","source":"agents/reviewer.toml","ownership":"normalized-agent","sha256":"d68c472fab6aef7679330798d753b3c495515545522b6f0a54cabccc0f523c39","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/SKILL.md","source":"skill/SKILL.md","ownership":"exact","sha256":"d32bcb8c79e21403bfdce45ede78719b5c992d1a7b9b0064327212dad7c35691","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/data/contracts/orchestration-v1.schema.json","source":"skill/data/contracts/orchestration-v1.schema.json","ownership":"exact","sha256":"0e8d5ebb06b47414be1218a1fc93cdbba94a7300fc7ba1c0e12d9c1ca60aeb43","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/data/policies/workflow-migrations-v1.json","source":"skill/data/policies/workflow-migrations-v1.json","ownership":"exact","sha256":"ce3786b60d65e4502c465ad33c2e6e105f996b067a315d382bf524797357644a","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/data/templates/ENGINEERING.md","source":"skill/data/templates/ENGINEERING.md","ownership":"exact","sha256":"3cd9dd02232897043ed9ff73e5076817dc4660696c8dfc8b06f4e5bad3158aad","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/data/templates/bridges/AGENTS.md","source":"skill/data/templates/bridges/AGENTS.md","ownership":"exact","sha256":"d32e8d147dd5f75cc63915bd0fe18bd4b0dd400abc8765e6986f5aed127dd585","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/data/templates/tools/ap.py","source":"skill/data/templates/tools/ap.py","ownership":"exact","sha256":"c352fbf2cbcb9692a34fd952d8541b1cbe6d475c6e11e64500a1f6b5283a00be","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/requirements.txt","source":"skill/requirements.txt","ownership":"exact","sha256":"0969da99a0bc2a1b71ed50584560f4588a37567ac63af3ddbaf3c4617ca5621a","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/scripts/ap.py","source":"skill/scripts/ap.py","ownership":"exact","sha256":"a8c157b3590dbd7619c1b8e99188b118174cc25d3a2162b598b7db01093f9fb1","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/scripts/core.py","source":"skill/scripts/core.py","ownership":"exact","sha256":"640c37e8b274d90d6b7765d4243e054360d5a5a85b3150fbfe6449527c1cd28f","executable":true,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/scripts/http_checks.py","source":"skill/scripts/http_checks.py","ownership":"exact","sha256":"5620e187a0129f135621425a84b3a4e8f31137263c7e549a36a801fc36ff54ea","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/scripts/install_integrity.py","source":"skill/scripts/install_integrity.py","ownership":"exact","sha256":"a9ca7bd67893bbd2d7b35ece30a40457f0cbcb9ae01580cfc5daca2cd96fd676","executable":false,"scope":"shared","version":"4.2.2"},{"path":".agents/skills/auto-coding-skill/scripts/scaffold_templates.py","source":"skill/scripts/scaffold_templates.py","ownership":"exact","sha256":"6f0ee92eff2d5db9627a64b332738c9c917a93cc559d5bbacf771c5966215821","executable":false,"scope":"shared","version":"4.2.2"},{"path":"AGENTS.md","source":"skill/data/templates/bridges/AGENTS.md","ownership":"exact","sha256":"d32e8d147dd5f75cc63915bd0fe18bd4b0dd400abc8765e6986f5aed127dd585","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/ENGINEERING.md","source":"skill/data/templates/ENGINEERING.md#managed-workflow","ownership":"managed-workflow","sha256":"c6bbd049e3939e503238b08a7f933bb371d1f63cf16aaab737925f4208563679","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/architecture/adr/_TEMPLATE-ADR.md","source":"generated:scaffold_templates:docs/architecture/adr/_TEMPLATE-ADR.md","ownership":"exact","sha256":"88c89065ae0f8555412dbe096c05342152efc7becae8d8981e6442cdfee8e672","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/deployment/deploy-records/_TEMPLATE-DEPLOY-RECORD.md","source":"generated:scaffold_templates:docs/deployment/deploy-records/_TEMPLATE-DEPLOY-RECORD.md","ownership":"exact","sha256":"f2c17c0febc322b0ad50fb9d6f7e64e6c0cfb15c8a3e445a99da40f0a0dd1901","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/design/_TEMPLATE-DD.md","source":"generated:scaffold_templates:docs/design/_TEMPLATE-DD.md","ownership":"exact","sha256":"6ca022d127bfff68521f466b60bace4f1206baf0913daa855b9f751e74198db3","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/reviews/_TEMPLATE-REVIEW.md","source":"generated:scaffold_templates:docs/reviews/_TEMPLATE-REVIEW.md","ownership":"exact","sha256":"134f0d1bd3c7a317ef3cf4c2aea9f0539f3d53dccd5bcdabf6915b3629ef8802","executable":false,"scope":"project","version":"4.2.2"},{"path":"docs/tools/autopipeline/ap.py","source":"skill/data/templates/tools/ap.py","ownership":"exact","sha256":"c352fbf2cbcb9692a34fd952d8541b1cbe6d475c6e11e64500a1f6b5283a00be","executable":false,"scope":"project","version":"4.2.2"}],"managed_namespaces":[{"path":".agents/skills/auto-coding-skill","scope":"shared"},{"path":"docs/tools/autopipeline","scope":"project"}],"preserved":[".agents/archive/**",".agents/agents/* (except paths listed in entries)","docs/project/**","docs/architecture/structure-standard.md","docs/architecture/adr/[0-9]*.md","docs/design/*.md","docs/interfaces/*.md","docs/deployment/deploy-records/*.md","docs/reviews/*.md"]}
|
|
@@ -27,24 +27,21 @@ python3 docs/tools/autopipeline/ap.py classify --scope auto \
|
|
|
27
27
|
[--task-kind read_only|change|terminal_maintenance] [--claim-direct]
|
|
28
28
|
```
|
|
29
29
|
|
|
30
|
-
Run `mechanism_plan.required
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
30
|
+
Run `mechanism_plan.required`; select `optional_when_beneficial` only when its
|
|
31
|
+
value exceeds coordination cost, and keep `forbidden` off absent user override.
|
|
32
|
+
`classify` is a fail-closed snapshot: inspect `repo`, `workspace_dirty`,
|
|
33
|
+
`dirty_paths`, and `active_writer`, and reclassify if state changes before writing.
|
|
34
|
+
Same-owner claims conflict unless continuing the exact ID; `change_nature` never bypasses a rule or review.
|
|
35
35
|
|
|
36
|
-
Use the registered lifecycle only when classification requires isolation or
|
|
37
|
-
fingerprinted review, or when the user explicitly requests it:
|
|
36
|
+
Use the registered lifecycle only when classification requires isolation/fingerprinted review or the user explicitly requests it:
|
|
38
37
|
|
|
39
38
|
```bash
|
|
40
39
|
python3 docs/tools/autopipeline/ap.py task-start T0001 \
|
|
41
40
|
--owned-path src [--isolated] [--review-required] [--continue-direct --direct-claim <ID>]
|
|
42
41
|
# Before the first commit, expand an active task instead of restarting it:
|
|
43
42
|
python3 docs/tools/autopipeline/ap.py task-scope-add T0001 --owned-path config/new-scope
|
|
44
|
-
#
|
|
45
|
-
python3 docs/tools/autopipeline/ap.py
|
|
46
|
-
python3 docs/tools/autopipeline/ap.py task-review T0001 \
|
|
47
|
-
--verdict approved --diff-fingerprint <CURRENT_DIFF_FINGERPRINT> --reviewer <REVIEWER_ID>
|
|
43
|
+
# Run a read-only Reviewer under its fixed deadline and record the bound result:
|
|
44
|
+
python3 docs/tools/autopipeline/ap.py review-run T0001 --reviewer <REVIEWER_ID> --json
|
|
48
45
|
python3 docs/tools/autopipeline/ap.py commit-push T0001 --msg "T0001: summary"
|
|
49
46
|
# Isolated tasks only:
|
|
50
47
|
python3 docs/tools/autopipeline/ap.py task-integrate T0001
|
|
@@ -52,7 +49,16 @@ python3 docs/tools/autopipeline/ap.py task-integrate T0001
|
|
|
52
49
|
|
|
53
50
|
Each parallel writer uses a task ID, worktree, lease, dependency SHAs, and distinct
|
|
54
51
|
`owned_paths`. Main integrates, gates, pushes, and cleans. Validate delegated JSON
|
|
55
|
-
with `agent-contract-check`; parallel fixers require reclassification with
|
|
52
|
+
with `agent-contract-check`; parallel fixers require reclassification with
|
|
53
|
+
`--writers >1`.
|
|
54
|
+
|
|
55
|
+
Use `reviewer` at `xhigh`: focused gets 90 seconds; parallel/cross-module or sensitive
|
|
56
|
+
boundaries get 300. `review-run` creates the assignment, launches a separate read-only
|
|
57
|
+
Codex process without the lifecycle-owner identity, terminates its process group at the
|
|
58
|
+
deadline, and records only an exact HEAD/scope/fingerprint match. Timeout is `blocked`.
|
|
59
|
+
`review-assignment` alone requires another deadline-capable host and cannot stop an
|
|
60
|
+
in-app subagent. `agent-result-template` supplies all 16 fields; `review-run` safely
|
|
61
|
+
normalizes presentation fields from the same analysis.
|
|
56
62
|
|
|
57
63
|
## Close with the bounded routed gate
|
|
58
64
|
|
|
@@ -71,7 +77,8 @@ Every changed code/config path must map to a real project command. The runtime
|
|
|
71
77
|
de-duplicates commands and enforces route, command, and total budgets. Defaults
|
|
72
78
|
are 120 seconds per command and 180 total; projects may raise them for a measured
|
|
73
79
|
affected-scope check. Timeouts should narrow the route, never trigger a full build.
|
|
74
|
-
Focused test/fix/retest loops remain allowed before the final gate.
|
|
80
|
+
Focused test/fix/retest loops remain allowed before the final gate. Commands inherit
|
|
81
|
+
the caller's PATH in non-login Bash; declare activation in the configured command.
|
|
75
82
|
|
|
76
83
|
Run full regression, Docker, Jenkins, deployment, browser/device writes, or target
|
|
77
84
|
acceptance only as an explicit diagnostic. A requested just-pushed failure
|
|
@@ -88,11 +95,8 @@ autocoding init
|
|
|
88
95
|
python3 docs/tools/autopipeline/ap.py doctor
|
|
89
96
|
```
|
|
90
97
|
|
|
91
|
-
Finish registered tasks before changing versions. `autocoding init`
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
and deploy-record artifacts in their designated directories. It then verifies the
|
|
97
|
-
packaged managed-install manifest; `doctor` cheaply rechecks the same declared
|
|
98
|
-
paths without treating project-owned files or archives as a release mirror.
|
|
98
|
+
Finish registered tasks before changing versions. `autocoding init` idempotently
|
|
99
|
+
replaces managed constraints, preserves schema-supported values, installs canonical
|
|
100
|
+
docs/templates, archives obsolete content outside active `docs/`, and preserves
|
|
101
|
+
designated artifacts. It verifies the managed-install manifest; `doctor` cheaply
|
|
102
|
+
rechecks declared paths without treating project-owned files or archives as a mirror.
|
|
@@ -124,6 +124,28 @@
|
|
|
124
124
|
"owning_fixer": {
|
|
125
125
|
"type": "string",
|
|
126
126
|
"minLength": 1
|
|
127
|
+
},
|
|
128
|
+
"review_depth": {
|
|
129
|
+
"enum": [
|
|
130
|
+
"focused",
|
|
131
|
+
"deep"
|
|
132
|
+
]
|
|
133
|
+
},
|
|
134
|
+
"timeout_seconds": {
|
|
135
|
+
"type": "integer",
|
|
136
|
+
"minimum": 1
|
|
137
|
+
},
|
|
138
|
+
"issued_at": {
|
|
139
|
+
"type": "string",
|
|
140
|
+
"minLength": 1
|
|
141
|
+
},
|
|
142
|
+
"deadline_at": {
|
|
143
|
+
"type": "string",
|
|
144
|
+
"minLength": 1
|
|
145
|
+
},
|
|
146
|
+
"scope_revision": {
|
|
147
|
+
"type": "integer",
|
|
148
|
+
"minimum": 1
|
|
127
149
|
}
|
|
128
150
|
},
|
|
129
151
|
"allOf": [
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
---
|
|
2
2
|
workflow:
|
|
3
|
-
skill_version: "4.2.
|
|
3
|
+
skill_version: "4.2.2"
|
|
4
4
|
mode: "dev"
|
|
5
5
|
profile: "auto"
|
|
6
6
|
completion: "push"
|
|
@@ -82,7 +82,7 @@ docs:
|
|
|
82
82
|
api_change_log: "docs/interfaces/api-change-log.md"
|
|
83
83
|
---
|
|
84
84
|
# Engineering Configuration and Project Facts
|
|
85
|
-
<!-- auto-coding-skill:managed-workflow:start version=4.2.
|
|
85
|
+
<!-- auto-coding-skill:managed-workflow:start version=4.2.2 -->
|
|
86
86
|
Root `AGENTS.md` is the single shared behavioral protocol. The installed
|
|
87
87
|
`SKILL.md` contains invocation guidance. This file is the exact-schema source for
|
|
88
88
|
project workflow configuration and access values that neither file can infer.
|
|
@@ -93,7 +93,7 @@ The frontmatter contract is:
|
|
|
93
93
|
- `commands`: project-native executable commands.
|
|
94
94
|
- `validation`: explicit path-to-command routes plus the bounded final-gate time
|
|
95
95
|
budget. Code/config paths must be mapped; route-level `timeout_seconds` can
|
|
96
|
-
bound a slower command
|
|
96
|
+
bound a slower command; commands inherit PATH and declare intentional activation.
|
|
97
97
|
- `risk`: project-specific signals that increase design/review depth but never
|
|
98
98
|
expand the automatic local gate.
|
|
99
99
|
- `structure` and `optimization`: advisory architecture and no-new-debt policy.
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
<!-- auto-coding-skill:managed-agents:start version=4.2.
|
|
1
|
+
<!-- auto-coding-skill:managed-agents:start version=4.2.2 -->
|
|
2
2
|
# Shared Engineering Protocol
|
|
3
3
|
|
|
4
4
|
This file is fully managed by `auto-coding-skill`. Keep project-specific facts,
|
|
@@ -34,17 +34,15 @@ The default budget is:
|
|
|
34
34
|
| High-risk or contract-crossing change | necessary design and independent review |
|
|
35
35
|
| Multiple independent writers | one task/worktree/lease per writer, ordered integration |
|
|
36
36
|
|
|
37
|
-
Do not add classify,
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
37
|
+
Do not add classify, lifecycle, design, subagents, review, or a worktree merely
|
|
38
|
+
because it exists. If task kind or impact is unclear, run `classify`, obey required
|
|
39
|
+
and forbidden mechanisms, and select optional ones by expected value. Classification
|
|
40
|
+
uses one fail-closed `git status --porcelain=v2 -z --untracked-files=all --no-renames`
|
|
41
|
+
snapshot; inspect `repo`, `workspace_dirty`, `dirty_paths`, and `active_writer`.
|
|
42
|
+
Git errors/malformed status block, and state changes before writing require reclassification.
|
|
43
43
|
|
|
44
|
-
Normal delivery is
|
|
45
|
-
|
|
46
|
-
`analysis → decomposition → necessary design → development → one bounded final
|
|
47
|
-
changed-scope gate → commit/push`.
|
|
44
|
+
Normal delivery is `analysis → decomposition → necessary design → development →
|
|
45
|
+
one bounded final changed-scope gate → commit/push`.
|
|
48
46
|
|
|
49
47
|
Push to the target branch ends normal coding. Do not poll Jenkins, deploy, or run
|
|
50
48
|
owner acceptance automatically. A user-requested diagnosis of the just-pushed
|
|
@@ -52,8 +50,8 @@ failure may continue in the same conversation/task without a second lifecycle.
|
|
|
52
50
|
|
|
53
51
|
## Git and parallel work
|
|
54
52
|
|
|
55
|
-
- One writer in a clean checkout works directly.
|
|
56
|
-
claim
|
|
53
|
+
- One writer in a clean checkout works directly. Claim uncertain direct work before
|
|
54
|
+
writing; any other claim, including the same owner, conflicts unless continuing its exact ID.
|
|
57
55
|
- Existing unrelated changes, another writer, or configured mandatory isolation
|
|
58
56
|
requires a registered task branch/worktree.
|
|
59
57
|
- Every delegated fixer or parallel writer owns a task ID/worktree, writer lease,
|
|
@@ -81,6 +79,16 @@ failure may continue in the same conversation/task without a second lifecycle.
|
|
|
81
79
|
- Validate delegated assignment/result JSON with `agent-contract-check` before use.
|
|
82
80
|
- Require fingerprinted review only for path/rule-confirmed high-risk,
|
|
83
81
|
contract-crossing, parallel, or configured work; intent words are candidates only.
|
|
82
|
+
- Keep `reviewer` at `xhigh`: focused is one 90-second analysis; deep gets 300
|
|
83
|
+
seconds for parallel/cross-module or sensitive boundaries. Timeout is `blocked`
|
|
84
|
+
and is not sent to `task-review`.
|
|
85
|
+
- Run `python3 docs/tools/autopipeline/ap.py review-run <TASK> --reviewer <ID> --json`.
|
|
86
|
+
It launches a separate read-only Codex process without the lifecycle-owner identity,
|
|
87
|
+
stops its process group at the fixed deadline, and records only the assigned HEAD,
|
|
88
|
+
scope, identity, and fingerprint. `review-assignment` alone requires a different
|
|
89
|
+
deadline-capable host and cannot stop an in-app subagent.
|
|
90
|
+
- `agent-result-template` supplies all 16 fields; contract checking aggregates errors,
|
|
91
|
+
and `review-run` safely normalizes presentation fields from the same analysis.
|
|
84
92
|
- Review blocks only defects introduced or worsened in the promised scope.
|
|
85
93
|
Adjacent existing issues are non-blocking follow-ups.
|
|
86
94
|
- Semantic changes invalidate approval. Mechanical documentation-only corrections
|