@elvis1513/auto-coding-skill 4.1.0 → 4.1.2

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.
@@ -1,135 +1,123 @@
1
- <!-- auto-coding-skill:managed-agents:start version=4.1.0 -->
2
-
3
- # Repository Delivery Rules
1
+ <!-- auto-coding-skill:managed-agents:start version=4.1.2 -->
2
+ # Shared Engineering Protocol
4
3
 
5
4
  This file is fully managed by `auto-coding-skill`. Keep project-specific facts,
6
- commands, access values, path mappings, architecture boundaries, and runtime
7
- notes in `docs/ENGINEERING.md` or its linked project documents. Do not add local
8
- workflow rules here; the next Skill sync replaces this file as a whole.
5
+ access values, risk rules, and validation routes in `docs/ENGINEERING.md`.
6
+
7
+ ## Authority
9
8
 
10
- ## Required sources
9
+ Use this order when sources disagree:
11
10
 
12
- Before changing a repository, read:
11
+ 1. The current user request.
12
+ 2. Code, tests, schemas, migrations, and runtime configuration for behavior that
13
+ exists now.
14
+ 3. `docs/ENGINEERING.md` for project facts and workflow configuration.
15
+ 4. Current interface documentation for intended contracts.
16
+ 5. Accepted DD/ADR for lasting decisions.
17
+ 6. Taskbooks, closures, reviews, and deployment records as history only.
13
18
 
14
- 1. the current user request;
15
- 2. this file;
16
- 3. `docs/ENGINEERING.md` frontmatter and project-specific sections;
17
- 4. only the source, tests, contracts, migrations, runtime configuration, or
18
- decision records relevant to the requested scope.
19
+ Update authoritative documentation when a change intentionally changes its
20
+ contract. Never let stale history override current implementation.
19
21
 
20
- Use this authority order when sources disagree:
22
+ ## Minimum mechanism budget
21
23
 
22
- 1. the current user request;
23
- 2. executable code, tests, schemas, migrations, and runtime configuration for
24
- behavior that exists now;
25
- 3. `docs/ENGINEERING.md` for project boundaries, access, risk, and validation;
26
- 4. interface documentation for intended external contracts;
27
- 5. accepted DD/ADR documents for lasting decisions;
28
- 6. taskbooks, closure logs, summaries, reviews, and deployment records as
29
- historical evidence only.
24
+ Use only mechanisms that improve expected delivery speed or defect prevention.
25
+ The default budget is:
30
26
 
31
- Do not let stale historical documents override current implementation. If a
32
- requested change intentionally changes a contract or decision, update the
33
- authoritative documentation in the same change.
27
+ | Situation | Required mechanisms |
28
+ | --- | --- |
29
+ | Read-only question | analysis only; no workflow command |
30
+ | Obvious clean serial edit | analysis, affected gate, commit/push |
31
+ | Terminal ledger/archive maintenance | targeted consistency check, one commit |
32
+ | Dirty checkout or concurrent writer | registered lifecycle and isolated worktree |
33
+ | High-risk or contract-crossing change | necessary design and independent review |
34
+ | Multiple independent writers | one task/worktree/lease per writer, ordered integration |
34
35
 
35
- ## Delivery flow
36
+ Do not add classify, task lifecycle, durable design, subagents, reviewer, or a
37
+ worktree merely because the tool exists. If task kind or impact is unclear, run
38
+ `classify`: run `mechanism_plan.required`, let the model select
39
+ `optional_when_beneficial` only when expected value exceeds coordination cost,
40
+ and keep `forbidden` mechanisms off unless the user explicitly overrides the
41
+ plan. Reclassify only after a material task-kind, scope, risk, or writer change.
36
42
 
37
- This Skill is a selectable guardrail, not a mandatory ceremony. The model first
38
- decides which mechanisms have positive net value for the current task. Read-only
39
- work, an obvious clean-checkout edit, or terminal ledger reconciliation normally
40
- needs no classify command, task lifecycle, design record, reviewer, or subagent.
41
- Use the smallest safe path for repository changes:
43
+ Normal delivery is:
42
44
 
43
- `analysis → decomposition → necessary design → development → one final
45
+ `analysis → decomposition → necessary design → development → one bounded final
44
46
  changed-scope gate → commit/push`.
45
47
 
46
- - Analyze impact and reuse points before writing.
47
- - Split only genuinely independent work; keep dependencies and file ownership
48
- explicit.
48
+ Push to the target branch ends normal coding. Do not poll Jenkins, deploy, or run
49
+ owner acceptance automatically. A user-requested diagnosis of the just-pushed
50
+ failure may continue in the same conversation/task without a second lifecycle.
51
+
52
+ ## Git and parallel work
53
+
54
+ - One writer in a clean checkout works directly on the current branch without a
55
+ task lifecycle.
56
+ - Existing unrelated changes, another writer, or configured mandatory isolation
57
+ requires a registered task branch/worktree.
58
+ - Every writing agent owns a distinct task ID, worktree, writer lease,
59
+ non-overlapping `owned_paths`, and prerequisite commit SHAs.
60
+ - Never let multiple writing agents share a checkout. Never restore, reset, stash,
61
+ clean, overwrite, or commit unknown changes.
62
+ - No diff means no commit, push, or temporary branch.
63
+ - The main agent integrates in dependency order and removes only clean, safely
64
+ merged temporary worktrees and branches.
65
+ - Do not sync or upgrade while any registered task is active.
66
+
67
+ ## Design, agents, and review
68
+
69
+ - Micro and standard tasks stay main-agent-only unless the model identifies an
70
+ independent question whose latency or expertise benefit exceeds coordination
71
+ cost.
49
72
  - Create DD/ADR only for lasting cross-module, API, data, security, deployment,
50
73
  or key user-flow decisions.
51
- - Focused tests may run and rerun during implementation. Run the configured final
52
- routed gate once after the diff is stable.
53
- - Commit and push all intended changes. A successful target-branch push ends
54
- normal coding work.
55
- - Do not poll Jenkins, deploy, or perform target acceptance automatically. If the
56
- user explicitly asks to diagnose a failure caused by the pushed change, keep
57
- that diagnosis and fix in the same conversation and task scope without
58
- creating an artificial new ledger task.
59
-
60
- ## Git safety and parallel work
61
-
62
- Inspect the checkout before writing.
63
-
64
- - One writer in a clean checkout works on the current branch.
65
- - Pre-existing unrelated changes, an already active writer, or two or more
66
- writers require isolated task branches/worktrees.
67
- - Each parallel writer gets one registered worktree, non-overlapping
68
- `owned_paths`, dependency commit SHAs, a distinct task ID, and a writer lease.
69
- - Never let multiple writing agents share a checkout.
70
- - Never restore, reset, stash, clean, overwrite, or commit unknown changes.
71
- - If a task produces no diff, create no commit or push. Direct work creates no
72
- temporary branch.
73
- - The main agent alone integrates by dependency order, pushes the target branch,
74
- and removes only clean, safely merged temporary worktrees and branches.
75
-
76
- Do not upgrade or sync the Skill while any registered task is active. Finish,
77
- integrate, or clean it with the installed runtime first so lifecycle semantics do
78
- not change mid-task.
79
-
80
- ## Risk, agents, and review
81
-
82
- - Micro and standard tasks default to the main agent. Do not require design,
83
- fan-out, a task ledger, or a reviewer merely because a change is non-trivial.
84
- - Use read-only explorer, documentation researcher, or browser roles only for
85
- independent questions whose expected value exceeds coordination cost.
86
- - Use parallel fixers only for bounded, dependency-aware units with separate
87
- worktrees and non-overlapping paths.
88
- - Require independent review for high-risk, cross-module, parallel integration,
74
+ - Use read-only explorer/docs/browser agents only for independent questions.
75
+ - Use parallel fixers only for bounded, dependency-free, non-overlapping units.
76
+ - Require fingerprinted review only for high-risk, contract-crossing, parallel,
89
77
  or explicitly configured work.
90
- - Review blocks only defects introduced or worsened inside the promised task
91
- scope. Record adjacent pre-existing issues as non-blocking follow-ups.
92
- - Any semantic code/config change invalidates review approval and requires a new
93
- diff fingerprint. A mechanical documentation-only correction may receive a
94
- targeted recheck by the same reviewer against the new fingerprint.
95
- - Historical debt does not block ordinary delivery. Block only new or worsened
96
- P0/P1 issues; handle broad cleanup through an explicit governance task.
97
-
98
- ## Validation
99
-
100
- `risk.rules` controls reasoning and review depth. `validation.routes` controls
101
- executable checks. Keep them separate.
102
-
103
- - Every changed code/config path must match one or more explicit validation
104
- routes in `docs/ENGINEERING.md`.
105
- - Run every matched command once in stable configuration order and de-duplicate
106
- repeated command references.
107
- - Unmapped code or a missing/blank command fails before staging.
108
- - Documentation-only work may use the built-in diff/format check.
109
- - `git diff --check` is additional hygiene and never business-code validation.
110
- - Use project-native fast checks. Do not automatically run a full repository
111
- build, Docker, Jenkins, deployment, live-device writes, or target validation.
112
- - Do not install dependencies proactively. Only after the selected final route
113
- fails because a repository-locked dependency is absent may it be restored once;
114
- then retry only that affected route.
115
- - External-system or real-device writes require the project's explicit safety
116
- rule or user confirmation; prefer fixtures/simulators for repeatable local
117
- coverage where the project supplies them.
78
+ - Review blocks only defects introduced or worsened in the promised scope.
79
+ Adjacent existing issues are non-blocking follow-ups.
80
+ - Semantic changes invalidate approval. Mechanical documentation-only corrections
81
+ may receive a targeted recheck by the same reviewer.
82
+ - Historical debt does not block ordinary delivery; block only new or worsened
83
+ P0/P1 issues.
84
+
85
+ ## Bounded real validation
86
+
87
+ - `risk.rules` controls reasoning/review depth. `validation.routes` alone selects
88
+ executable closure checks.
89
+ - Every changed code/config path must match one or more explicit routes. Run every
90
+ matched command once in stable order and de-duplicate references.
91
+ - Unmapped code or blank commands fail before staging. Documentation-only work may
92
+ use the built-in diff/format check. `git diff --check` is hygiene, not business
93
+ validation.
94
+ - Focused tests may run and rerun during development. Run one final routed gate
95
+ after the diff is stable.
96
+ - The final gate defaults to 120 seconds per command and 180 seconds total. A
97
+ project may raise either budget for a measured affected-scope check, and may set
98
+ a smaller `timeout_seconds` on an individual route. A timeout should narrow the
99
+ route; never expand or retry it as a full gate.
100
+ - Do not proactively install dependencies. Only after the selected route fails
101
+ because a repository-locked dependency is absent may it be restored once and
102
+ only that route retried.
103
+ - Full regression, repository-wide scans, builds, Docker, Jenkins, deployment,
104
+ live-device writes, browser/API acceptance, and target checks are explicit
105
+ diagnostics, not normal closure.
106
+ - External-system writes require the project safety rule or user confirmation;
107
+ use fixtures/simulators when the project supplies them.
118
108
 
119
109
  ## Documentation and access
120
110
 
121
- - `docs/ENGINEERING.md` is the single project workflow configuration source. Its
122
- managed block explains the shared protocol; content outside that block contains
123
- project facts, not competing workflow instructions.
124
- - Put durable product/domain facts, repository maps, and runtime/deployment facts
125
- in `docs/project/` when they are too large for `ENGINEERING.md`.
126
- - Ordinary work does not create taskbook, closure Markdown, evidence JSONL,
127
- active-task documents, or design files. Pure ledger/archive reconciliation is
128
- a terminal maintenance action and must not create another task lifecycle.
129
- - Generated task state, leases, fingerprints, and gate evidence stay in Git
130
- common/local state so recording them cannot change the reviewed diff.
131
- - Fill the configured project/Jenkins/GitLab/Nexus access fields during project
132
- initialization. Plaintext values are allowed by the generic workflow; do not
133
- invent or echo credentials unnecessarily.
111
+ - This AGENTS file is the single behavioral protocol. The installed `SKILL.md`
112
+ contains invocation guidance. `docs/ENGINEERING.md` is the single project
113
+ configuration and facts source. Do not repeat the protocol elsewhere.
114
+ - Put larger product, repository, and runtime facts under `docs/project/` and link
115
+ them from ENGINEERING.
116
+ - Ordinary work creates no taskbook, closure Markdown, evidence JSONL, active-task
117
+ document, or design file. Machine coordination/evidence stays in Git
118
+ common/local state and cannot change the reviewed diff.
119
+ - Pure ledger/archive reconciliation is terminal: validate, commit once, and stop.
120
+ - Fill configured project/Jenkins/GitLab/Nexus access fields during initialization.
121
+ Plaintext values are allowed; do not invent or unnecessarily echo credentials.
134
122
 
135
123
  <!-- auto-coding-skill:managed-agents:end -->