vibe-coding-master 0.4.41 → 0.5.0

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.
@@ -1,160 +0,0 @@
1
- # Full Harness Baseline
2
-
3
- Last updated: 2026-06-21
4
-
5
- Status: Current implementation reference.
6
-
7
- This file tracks the VCM fixed harness currently installed by
8
- `src/backend/services/harness-service.ts` and
9
- `src/backend/cli/install-vcm-harness.ts`.
10
-
11
- ## Principles
12
-
13
- VCM separates three file classes:
14
-
15
- - Harness-managed files: VCM installs, upgrades, repairs, and can overwrite
16
- inside managed blocks or whole-file templates.
17
- - Project-owned durable docs: VCM may create starter docs, but project roles own
18
- their content afterward.
19
- - Runtime state: VCM writes these during task execution and cleanup removes them
20
- when the task is closed.
21
-
22
- The current implementation no longer uses `.ai/vcm-harness-manifest.json`.
23
- Harness ownership is defined by the deterministic installer and by managed
24
- markers such as `<!-- VCM:BEGIN version=1 -->`.
25
-
26
- ## Fixed Harness Files
27
-
28
- Managed-block files:
29
-
30
- ```text
31
- CLAUDE.md
32
- .gitignore
33
- .github/pull_request_template.md
34
- .claude/agents/project-manager.md
35
- .claude/agents/architect.md
36
- .claude/agents/coder.md
37
- .claude/agents/reviewer.md
38
- .claude/agents/gate-reviewer.md
39
- .claude/agents/translator.md
40
- ```
41
-
42
- Whole-file or raw-file harness files:
43
-
44
- ```text
45
- .claude/skills/vcm-route-message/SKILL.md
46
- .claude/skills/vcm-final-acceptance/SKILL.md
47
- .claude/skills/vcm-long-running-validation/SKILL.md
48
- .claude/skills/vcm-harness-bootstrap/SKILL.md
49
- .claude/skills/vcm-gate-review/SKILL.md
50
- .claude/agents/translator.md
51
- .ai/tools/generate-module-index
52
- .ai/tools/generate-public-surface
53
- .ai/tools/request-gate-review
54
- .ai/tools/run-long-check
55
- .ai/tools/watch-job
56
- .ai/tools/vcm-bash-guard
57
- ```
58
-
59
- Durable doc templates created only when missing:
60
-
61
- ```text
62
- docs/ARCHITECTURE.md
63
- docs/TESTING.md
64
- docs/known-issues.md
65
- ```
66
-
67
- Directory roots created by the installer:
68
-
69
- ```text
70
- .claude/agents/
71
- .claude/skills/
72
- .claude/skills/vcm-final-acceptance/
73
- .claude/skills/vcm-harness-bootstrap/
74
- .claude/skills/vcm-long-running-validation/
75
- .claude/skills/vcm-route-message/
76
- .claude/skills/vcm-gate-review/
77
- .ai/vcm/translations/
78
- .ai/vcm/gate-reviews/
79
- .ai/tools/
80
- .ai/generated/
81
- ```
82
-
83
- ## Installed Skills
84
-
85
- - `vcm-route-message`: route-file write protocol.
86
- - `vcm-final-acceptance`: PM final evidence audit.
87
- - `vcm-long-running-validation`: foreground supervision protocol for long jobs.
88
- - `vcm-harness-bootstrap`: AI-assisted project understanding and generated
89
- context refresh.
90
- - `vcm-gate-review`: PM protocol for requesting Gate Review and handling VCM
91
- callbacks.
92
-
93
- ## Installed Tools
94
-
95
- ```text
96
- .ai/tools/generate-module-index
97
- .ai/tools/generate-public-surface
98
- .ai/tools/request-gate-review
99
- .ai/tools/run-long-check
100
- .ai/tools/watch-job
101
- .ai/tools/vcm-bash-guard
102
- ```
103
-
104
- `generate-module-index` and `generate-public-surface` support Rust/Cargo
105
- projects and npm workspace TypeScript/JavaScript projects. Other repository
106
- shapes can still install the fixed harness, but generated context should be
107
- treated as unsupported until project-specific generators exist.
108
-
109
- ## Runtime State
110
-
111
- Task runtime state lives in the task worktree:
112
-
113
- ```text
114
- <taskRepoRoot>/.ai/vcm/sessions/<task>.json
115
- <taskRepoRoot>/.ai/vcm/messages/<task>.jsonl
116
- <taskRepoRoot>/.ai/vcm/orchestration/<task>.json
117
- <taskRepoRoot>/.ai/vcm/translation/<task>/
118
- <taskRepoRoot>/.ai/vcm/handoffs/
119
- <taskRepoRoot>/.ai/vcm/handoffs/messages/<from-role>-<to-role>.md
120
- <taskRepoRoot>/.ai/vcm/handoffs/architecture-plan.md
121
- <taskRepoRoot>/.ai/vcm/handoffs/known-issues.md
122
- <taskRepoRoot>/.ai/vcm/handoffs/review-report.md
123
- <taskRepoRoot>/.ai/vcm/handoffs/docs-sync-report.md
124
- <taskRepoRoot>/.ai/vcm/handoffs/final-acceptance.md
125
- <taskRepoRoot>/.ai/vcm/jobs/<job-id>/
126
- <taskRepoRoot>/.ai/vcm/gate-reviews/
127
- ```
128
-
129
- Project-scoped runtime state lives in the base repository:
130
-
131
- ```text
132
- <baseRepoRoot>/.ai/vcm/harness-engineer/session.json
133
- <baseRepoRoot>/.ai/vcm/translations/
134
- <baseRepoRoot>/.ai/vcm/bootstrap/session.json
135
- ```
136
-
137
- App-local state lives under `vcmDataDir`:
138
-
139
- ```text
140
- <vcmDataDir>/settings.json
141
- <vcmDataDir>/projects/index.json
142
- <vcmDataDir>/projects/<project-id>/config.json
143
- <vcmDataDir>/projects/<project-id>/tasks/<task>.json
144
- <vcmDataDir>/gateway/
145
- ```
146
-
147
- Ordinary VCM role sessions no longer persist raw terminal logs under
148
- `.ai/vcm/handoffs/logs/`. Claude transcript JSONL files under
149
- `~/.claude/projects/` remain the semantic source for output translation and
150
- session recovery.
151
-
152
- ## Cleanup
153
-
154
- Close Task removes the task-owned worktree, branch, app-local task record, and
155
- task-local runtime state. It stops task-scoped VCM role sessions for that task.
156
- It must not stop project-scoped Translator or Harness Engineer sessions.
157
-
158
- Durable facts that should survive task cleanup must be promoted into code,
159
- tests, durable docs, PR text, commit history, or release notes before closing
160
- the task.
@@ -1,132 +0,0 @@
1
- # Gate Review Gates
2
-
3
- This document defines VCM Gate Review Gates.
4
-
5
- Gate Review is an optional quality layer for complex tasks. It uses a fifth
6
- VCM flow role, `gate-reviewer`, powered by Claude Code. The role uses the same
7
- hook, Round, terminal, and translation path as the normal VCM roles, but it
8
- does not participate in PM route-file dispatch.
9
-
10
- ## Gates
11
-
12
- VCM supports three gates:
13
-
14
- - `architecture-plan`: review architect's plan before coder dispatch.
15
- - `validation-adequacy`: review reviewer validation coverage before docs sync
16
- or final acceptance.
17
- - `final-diff`: review the final diff before PR preparation.
18
-
19
- Each gate returns only:
20
-
21
- - `approve`
22
- - `request_changes`
23
-
24
- Gate Reviewer does not choose owners, fixes, Replan, or user intervention. PM
25
- routes follow-up by gate type.
26
-
27
- ## Ownership
28
-
29
- VCM owns:
30
-
31
- - global Gate Review switches in `~/.vcm/settings.json`
32
- - current task gate state under `<taskRepoRoot>/.ai/vcm/gate-reviews/`
33
- - starting/resuming the task-scoped Gate Reviewer Claude Code session
34
- - sending the gate prompt
35
- - waiting for the report
36
- - sending the callback to PM
37
-
38
- PM owns:
39
-
40
- - triggering `vcm-gate-review` at the three gate points
41
- - stopping after a `started` or `running` result
42
- - handling the VCM callback
43
- - routing `request_changes` to the responsible normal VCM role
44
-
45
- Gate Reviewer owns:
46
-
47
- - reviewing only the requested gate
48
- - writing only the assigned report
49
-
50
- ## Session Scope
51
-
52
- Gate Reviewer is task-scoped, like the normal VCM roles:
53
-
54
- ```text
55
- <taskRepoRoot>/<stateRoot>/sessions/<taskSlug>.json
56
- ```
57
-
58
- Each task gets its own Gate Reviewer session. The gate prompt names the current
59
- task and task worktree path explicitly, and only current worktree evidence can
60
- decide the gate.
61
-
62
- Gate reports remain task-scoped:
63
-
64
- ```text
65
- <taskRepoRoot>/.ai/vcm/gate-reviews/
66
- ```
67
-
68
- ## Prompt Shape
69
-
70
- VCM sends a short prompt:
71
-
72
- ```text
73
- [VCM GATE REVIEW]
74
- Task: <taskSlug>
75
- Worktree: <taskRepoRoot>
76
- Gate: <gate>
77
- Request: <requestId>
78
- Report: <absolute report path>
79
-
80
- Evidence:
81
- - <relative evidence file>
82
- Diff: inspect git status/diff in Worktree.
83
-
84
- Write only Report. Start exactly:
85
- Gate: <gate>
86
- Request: <requestId>
87
- Decision: approve|request_changes
88
- Summary: <one or two sentences>
89
- [/VCM GATE REVIEW]
90
- ```
91
-
92
- Detailed review rules live in `.claude/agents/gate-reviewer.md`, so the prompt
93
- does not repeat role policy.
94
-
95
- ## Files
96
-
97
- Harness-managed files:
98
-
99
- ```text
100
- .claude/agents/gate-reviewer.md
101
- .claude/skills/vcm-gate-review/SKILL.md
102
- .ai/tools/request-gate-review
103
- ```
104
-
105
- Task runtime files:
106
-
107
- ```text
108
- .ai/vcm/gate-reviews/index.json
109
- .ai/vcm/gate-reviews/architecture-plan-review.md
110
- .ai/vcm/gate-reviews/validation-adequacy-review.md
111
- .ai/vcm/gate-reviews/final-diff-review.md
112
- .ai/vcm/gate-reviews/requests/<request-id>.json
113
- .ai/vcm/gate-reviews/requests/<request-id>.prompt.md
114
- ```
115
-
116
- ## Flow
117
-
118
- ```text
119
- PM reaches gate
120
- -> PM uses vcm-gate-review
121
- -> VCM checks global gate switch and task state
122
- -> VCM starts or resumes the task Gate Reviewer session
123
- -> VCM sends short prompt with task/worktree/report paths
124
- -> Gate Reviewer writes report
125
- -> standard Claude hooks update Gate Reviewer activity/Round state
126
- -> VCM validates report
127
- -> VCM callbacks PM
128
- -> PM continues or routes follow-up
129
- ```
130
-
131
- Close Task stops the task-scoped Gate Reviewer session with the other VCM role
132
- sessions.