@su-record/vibe 2.9.3 → 2.9.4
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/package.json +1 -1
- package/skills/vibe.plan/SKILL.md +44 -6
package/package.json
CHANGED
|
@@ -55,13 +55,51 @@ Read .claude/vibe/interviews/{feature-name}.md
|
|
|
55
55
|
Extract `type`, `status`, `requiredCollected`, `optionalCollected`, etc. from the frontmatter.
|
|
56
56
|
|
|
57
57
|
**Validation**:
|
|
58
|
-
- `status: partial` + Required items incomplete → warn the user:
|
|
59
|
-
```
|
|
60
|
-
⚠️ Interview is partially complete (N Required items not collected).
|
|
61
|
-
Those items will appear as "TBD" in the plan document.
|
|
62
|
-
Continue? (y/N)
|
|
63
|
-
```
|
|
64
58
|
- If the file does not exist → guide the user to run `vibe.interview` first.
|
|
59
|
+
- `status: partial` + Required items incomplete → apply **AI-Driven Gap Filling**:
|
|
60
|
+
|
|
61
|
+
### AI-Driven Gap Filling
|
|
62
|
+
|
|
63
|
+
When the user stops the interview early, they are **delegating the rest to AI** — not refusing to answer. The plan phase must fill in the gaps autonomously.
|
|
64
|
+
|
|
65
|
+
**Process for missing Required items:**
|
|
66
|
+
|
|
67
|
+
1. **Analyze the project** — Read `package.json`, existing code, `CLAUDE.md`, `docs/` for context clues
|
|
68
|
+
2. **Research similar products** — Use the collected answers (purpose, target users) to infer reasonable defaults
|
|
69
|
+
3. **Fill with best-guess + rationale** — For each missing item, write the AI's best inference with a `[AI-inferred]` tag and a one-line rationale
|
|
70
|
+
|
|
71
|
+
**Example:**
|
|
72
|
+
|
|
73
|
+
```markdown
|
|
74
|
+
### Success Metrics
|
|
75
|
+
[AI-inferred] Based on the project type (landing page for coffee brand):
|
|
76
|
+
- Conversion rate: visitor → purchase click-through > 3%
|
|
77
|
+
- Page load: FCP < 1.5s on mobile
|
|
78
|
+
- Bounce rate: < 50%
|
|
79
|
+
```
|
|
80
|
+
|
|
81
|
+
4. **Present filled plan for review** — Show the user which items were AI-filled:
|
|
82
|
+
|
|
83
|
+
```
|
|
84
|
+
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
85
|
+
📄 Plan document generated
|
|
86
|
+
|
|
87
|
+
Collected from interview: {N} items
|
|
88
|
+
AI-inferred: {M} items (marked [AI-inferred])
|
|
89
|
+
|
|
90
|
+
Please review the [AI-inferred] sections.
|
|
91
|
+
Edit anything that doesn't match your intent.
|
|
92
|
+
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
93
|
+
```
|
|
94
|
+
|
|
95
|
+
5. **User can correct inline** — If the user says "change X to Y", update the plan and remove the `[AI-inferred]` tag from that item.
|
|
96
|
+
|
|
97
|
+
**Rules for AI inference:**
|
|
98
|
+
- Always choose the most common/safe default for the project type
|
|
99
|
+
- Never invent business requirements (e.g., pricing, brand name) — mark these as `[Needs user input]` instead
|
|
100
|
+
- Technical decisions (framework, DB, hosting) can be inferred from existing project setup
|
|
101
|
+
- UX decisions (color, tone, layout) can be inferred from reference sites or stated brand direction
|
|
102
|
+
- If fewer than 2 answers were collected (user barely started), ask 3 essential questions before proceeding: purpose, target users, and project type
|
|
65
103
|
|
|
66
104
|
### `.last-feature` Pointer Update
|
|
67
105
|
|