메뉴
Use when evaluation data exists and optimization proposals are needed. Analyzes scores, user journey drop-off points, and prior attempts to produce ranked hypotheses batched 2-3 per cycle.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/cbemister/claude-code-utils --skill generate-hypotheses명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Identify and fix accessibility issues for WCAG 2.1 Level AA compliance, including mobile screen reader support (VoiceOver/TalkBack), touch targets, and keyboard navigation.
Autonomously build an app from staged plans created by /launch-app. Reads plans/active/ stage files and builds each stage with verification, commits, and Slack notifications. Designed for unattended execution via build-app-runner.sh.
Polish components to production quality — interactive states, micro-interactions, and final refinement. Run all three passes or target one area.
SKILL.md 표시 중
SOC 직업 분류 기준
| name | generate-hypotheses |
| description | Use when evaluation data exists and optimization proposals are needed. Analyzes scores, user journey drop-off points, and prior attempts to produce ranked hypotheses batched 2-3 per cycle. |
| argument-hint | [count] - Optional: number of hypotheses to generate (default: 5, max: 10) |
Analyze evaluation data and produce ranked optimization proposals. Batches the top 2-3 hypotheses per evolution cycle for implementation.
CRITICAL — NON-INTERACTIVE: Execute all steps without pausing. Never ask questions, request confirmation, or wait for input. Read state, analyze, generate, write output, update state, stop.
Read factory/evolution-state.json:
"hypothesizing" or "rejected" (if rejected, we're retrying with next batch)CYCLE to current_cycleParse $ARGS for optional count (default: 5, max: 10)
Read the latest evaluation:
factory/evaluations/eval-*.jsonscores, top_issues, and user_journeysRead hypothesis history:
factory/hypotheses/hyp-*.jsonrejected_hypotheses from evolution stateFor each issue in top_issues, map it to a concrete improvement:
"small" (1-2 tasks), "medium" (3-5 tasks), "large" (6+ tasks)Additionally, analyze user journey drop-off points:
+2 to expected_impactFor each hypothesis, calculate priority:
effort_multiplier = { "small": 1, "medium": 2, "large": 3 }
priority_score = (expected_impact * confidence) / effort_multiplier
Apply modifiers:
expected_impact += 2 (already applied in Step 1)confidence *= 1.2 (cap at 1.0)confidence *= 0.5Sort by priority_score descending.
Select the top 2-3 hypotheses for this cycle:
A hypothesis conflicts if it modifies the same primary files as an already-batched hypothesis in a way that could cause merge issues.
For each batched hypothesis, produce an implementation sketch:
{
"pages_affected": ["src/app/page.tsx"],
"new_components": ["src/components/SocialProof.tsx"],
"modifications": ["src/app/layout.tsx"],
"estimated_tasks": 3,
"key_changes": [
"Add testimonial data to homepage",
"Create SocialProof component with carousel",
"Add trust badges below hero section"
]
}
Read the actual source files referenced to ensure the sketch is grounded in the real codebase.
Generate a timestamp: TIMESTAMP = current ISO-8601 datetime formatted as YYYY-MM-DD-HHmmss
Write factory/hypotheses/hyp-{CYCLE}-{TIMESTAMP}.json:
{
"version": 1,
"timestamp": "[ISO-8601]",
"cycle": 1,
"evaluation_ref": "factory/evaluations/eval-[latest].json",
"total_generated": 5,
"batched_count": 3,
"hypotheses": [
{
"id": "hyp-1-001",
"title": "Add social proof section below hero",
"category": "conversion",
"description": "Homepage lacks trust signals. Adding testimonials, logos, and stats below the hero would increase trust and reduce bounce rate.",
"rationale": "Conversion audit scored trust_credibility at 2/5. User journey 'New visitor' drops off before CTA due to lack of social validation.",
"expected_impact": 8,
"confidence": 0.8,
"effort": "small",
"effort_multiplier": 1,
"priority_score": 6.4,
"batched": true,
"addresses_issues": ["issue-1", "issue-3"],
"addresses_journeys": ["New visitor"],
"predicted_score_change": {
"conversion.trust_credibility": "+2",
"composite_score": "+5"
},
"implementation_sketch": {
"pages_affected": ["src/app/page.tsx"],
"new_components": ["src/components/SocialProof.tsx"],
"modifications": [],
"estimated_tasks": 3,
"key_changes": [
"Add testimonial data structure",
"Create SocialProof component",
"Integrate below hero on homepage"
]
}
}
],
"skipped": [
{
"id": "hyp-1-004",
"title": "...",
"reason": "Similar to rejected hypothesis hyp-0-002",
"priority_score": 3.2
}
]
}
Write factory/hypotheses/hyp-{CYCLE}-{TIMESTAMP}.md:
# Optimization Hypotheses — Cycle [N]
**Project:** [name]
**Date:** [date]
**Current Score:** [X]/100
**Evaluation:** [ref]
---
## Batched for This Cycle (2-3)
### 1. [Title] ⭐ Priority: [score]
- **Category:** [dimension]
- **Impact:** [expected_impact]/10 | **Confidence:** [confidence] | **Effort:** [effort]
- **Addresses:** [issue IDs], [journey names]
- **Predicted Score Change:** [composite delta]
- **Rationale:** [why this will work]
- **Key Changes:**
- [change 1]
- [change 2]
### 2. [Title] ...
---
## Other Candidates (Not Batched)
| Rank | Title | Priority | Reason Not Batched |
|------|-------|----------|-------------------|
| 4 | [title] | [score] | Conflicts with #1 |
| 5 | [title] | [score] | Lower priority |
---
## Skipped
| Title | Reason |
|-------|--------|
| [title] | Similar to rejected hyp-0-002 |
---
## Next: Run /plan-optimization to convert these into a stage plan
factory/evolution-state.json"planning"================================================================
Hypotheses Generated
================================================================
Project: [name]
Cycle: [N]
Generated: [total] hypotheses
Batched: [count] for this cycle
Batched Hypotheses:
1. [title] — priority [score], predicted +[X] score
2. [title] — priority [score], predicted +[X] score
3. [title] — priority [score], predicted +[X] score
Output:
JSON: factory/hypotheses/hyp-{CYCLE}-{TIMESTAMP}.json
Report: factory/hypotheses/hyp-{CYCLE}-{TIMESTAMP}.md
Next: Run /plan-optimization to create the stage plan
================================================================
| Scenario | Recovery |
|---|---|
| No evaluation files | Display error: "Run /evaluate-product first" |
| All hypotheses conflict | Batch only the top 1 |
| All hypotheses previously rejected | Generate from secondary issues, widen scope |
| Evaluation has no top_issues | Generate hypotheses from lowest-scoring dimensions |
| Context exhausted mid-generation | Runner re-invokes; check for partial output, resume |