Claw Research - Core Prompt

⛔ BLOCKER Rules (Must Pass)

Violation = Invalid Output - Must Redo

#	Rule	Check
1	Read sources from config/sources.json only	No hardcoded paths
2	Every evidence: full_text + source_url + source_type + published_at	4 fields required
3	Output analysis-result.json with valid JSON	Schema compliant
4	Present full Markdown report to user	No "see file" shortcuts
5	No truncated evidence text	Complete quotes
6	No placeholder text	No "TBD" or "..."

---

7-Stage Workflow

CHARTER → SOURCES → COLLECT → ANALYZE → MD_REPORT → NOTION → ACTION
 (opt)   [BLOCKER]   auto     AI core   [BLOCKER]  from MD   track

Core Principle: Markdown report is the "Single Source of Truth" (SSOT). Notion sync MUST read from MD file.

---

Stage 1: Charter (Optional)

Read workspace/projects/{project}/config/research-charter.json if exists.

---

Stage 2: Sources [BLOCKER]

Run pipeline to collect data:

python3 scripts/run_pipeline.py \
  --project-name "my-research" \
  --sources "hacker_news,woshipm,pmcaff" \
  --query "product pain points"

Key Options:

Option	Description	Default
--project-name	Unique project identifier	Required
--sources	Comma-separated sources	All enabled
--query	Search query	From charter
--lookback-hours	Time window	720 (30 days)
--max-items-per-source	Items per source	50
--include-keyword	Filter: must contain	None
--exclude-keyword	Filter: must not contain	None

Output: workspace/projects/{project}/data/candidate_items.jsonl

---

Stage 3: Collect (Automatic)

Collectors output to workspace/projects/{project}/data/:

• raw.jsonl - Raw data
• candidate_items.jsonl - Filtered items

---

Stage 4: Analyze (AI Core)

Read from data/:

• analysis_input.json
• candidate_items.jsonl
• research-charter.json (if exists)

AI Performs:

1. Pain point analysis
2. Payment signal detection
3. Bayesian probability scoring
4. Hypothesis validation (if charter)
5. Strategic analysis (SWOT, competitors, journey)

---

Stage 5: MD Report [BLOCKER]

Must generate Markdown file first as the single source for Notion sync.

1. Write JSON: workspace/projects/{project}/data/analysis-result.json
2. Write Markdown: workspace/projects/{project}/reports/analysis-report-{YYYY-MM-DD}.md
3. Present to user: Display full report in conversation

Markdown File Path:

workspace/projects/{project}/reports/analysis-report-{date}.md

⚠️ Important: This MD file is the ONLY data source for Notion sync. Must contain complete content.

---

Stage 6: Notion Sync (From MD File)

Read content from Markdown file to sync to Notion, ensuring consistency.

┌─────────────────────────────────────────────────────────┐
│  analysis-report-{date}.md  ──────►  Notion Page        │
│  (Single Source of Truth)         (Read from MD file)   │
└─────────────────────────────────────────────────────────┘

Sync Process:

1. Read complete content from reports/analysis-report-{date}.md
2. Parse Markdown structure (headings, tables, blockquotes, etc.)
3. Convert to Notion API format (callout, to_do, divider, etc.)
4. Create/Update Notion page

⚠️ Prohibited:

• ❌ Do NOT regenerate Notion content from analysis-result.json
• ❌ Do NOT skip MD file and sync directly
• ❌ Do NOT simplify or truncate content from MD file

Notion Sync Status:

## Notion Sync Status
- [x] Synced: https://notion.so/page-id
- [ ] SYNC_SKIPPED: [reason - e.g., "Notion MCP not configured"]

---

Stage 7: Action (Track)

Manage action items:

# List pending actions
python3 scripts/manage_actions.py --project "my-research" --list-pending

# Add new action
python3 scripts/manage_actions.py --project "my-research" \
  --add "Interview 5 target users" \
  --due "2026-04-15" \
  --insight "High-volume sellers show strongest pain"

# Complete action
python3 scripts/manage_actions.py --project "my-research" \
  --complete "ACTION-001" \
  --note "Completed 5 interviews, 3 confirmed willingness to pay"

---

Output Format

JSON Structure (Write to analysis-result.json)

{
  "generated_at": "UTC timestamp",
  "project_name": "string",
  "top_pain_points": [{ "label", "summary", "evidence_count", "example_urls" }],
  "candidate_clusters": [{ "name", "summary", "evidence_count", "payment_signal" }],
  "payment_signals": [{ "label", "summary" }],
  "bayesian_scores": {
    "success_probability": {
      "posterior": 0.35,
      "prior": { "value": 0.12, "reasoning": "string" },
      "confidence": "high|medium|low",
      "signal_assessments": [{ "dimension", "strength", "reasoning", "evidence_quotes", "contribution" }],
      "calculation_summary": "string",
      "key_uncertainties": ["string"]
    },
    "payment_probability": { /* same structure */ }
  },
  "hypothesis_validation": [{ "hypothesis_id", "status", "supporting_evidence", "missing_evidence" }],
  "strategic_analysis": {
    "swot": { "strengths", "weaknesses", "opportunities", "threats" },
    "competitor_landscape": [{ "name", "gap", "our_advantage" }],
    "user_journey_stages": [{ "stage", "pain_level", "evidence_count" }]
  },
  "actionable_insights": [{ "insight", "evidence_basis", "recommended_action", "priority" }],
  "action_items": [{ "id", "action", "owner", "status" }],
  "strongest_examples": [{ "title", "url", "reason" }],
  "risks_or_gaps": ["string"]
}

Markdown Report (Present to User)

# [Project] Market Research Report
**Generated**: YYYY-MM-DD | **Confidence**: high/medium/low

## Executive Summary
**Conclusion**: [One sentence]
**Actions**: 1. [P0] ... 2. [P1] ... 3. [P2] ...
**Risks**: [Top 2 risks]

## Bayesian Scores
| Probability | Prior | Posterior | Key Signals |
|-------------|-------|-----------|-------------|
| Success | 12% | 35% | pain_intensity +10% |
| Payment | 6% | 18% | economic_impact +8% |

## Top Pain Points
### Pain 1: [Title] (X evidence)
> **Original**: "[Full quote]"
> **Source**: [Name](URL) | **Type**: community_discussion | **Date**: YYYY-MM-DD

## Strategic Analysis
**SWOT**: Strengths: [evidence] | Weaknesses: [gap] | Opportunities: [trend] | Threats: [risk]
**Competitors**: [Competitor] - Gap: [gap] - Our Advantage: [ours]

## Action Items
- [ ] ACTION-001: [Task] (Due: YYYY-MM-DD)

## Notion Sync
[ ] Synced: [URL] OR [ ] SYNC_SKIPPED: [reason]

---

Bayesian Scoring

Formula

Posterior = Prior + Σ(Signal_Strength × Contribution)

Dimensions

Type	Dimensions
Success	pain_intensity, market_evidence, solution_gap, timing_signal, execution_fit
Payment	economic_impact, purchase_intent, budget_access, urgency, trust_signals

Scoring Rules

• Prior: Estimate from industry context (do NOT hardcode)
• Strength: 0.0-1.0 (based on evidence)
• Contribution: e.g., "+0.10"
• Confidence: high/medium/low

---

Evidence Format [BLOCKER]

Every evidence quote MUST include:

> **Original**: "[Complete original text]"
> **Source**: [Source Name](URL)
> **Type**: community_discussion
> **Published**: YYYY-MM-DD

Source Types: community_discussion | blog_article | industry_survey | competitor_review | documentation | interview | social_post

---

Completion Checklist

[BLOCKER] Must Pass

• analysis-result.json exists with valid JSON
• reports/analysis-report-{date}.md exists with full Markdown report
• Full Markdown report presented to user
• All evidence has: full_text + URL + type + date
• No truncated text, no placeholders

[REQUIRED] Should Pass

• Bayesian prior has reasoning
• Signal assessments have evidence_quotes
• key_uncertainties listed
• Notion sync completed FROM MD FILE (not from JSON)
• Notion sync status stated in conversation

[QUALITY] Consistency Check

• MD file content == User conversation output
• Notion content == MD file content (no simplification)
• Evidence quotes preserved across all outputs

---

File Structure

workspace/projects/{project}/
├── config/research-charter.json
├── data/
│   ├── raw.jsonl
│   ├── candidate_items.jsonl
│   ├── analysis_input.json
│   └── analysis-result.json
└── reports/
    ├── analysis-report-{date}.md   # ⭐ Single Source of Truth (SSOT)
    ├── {date}.md                   # Auto-generated brief report
    └── weekly-{year}-{week}.md     # Weekly report

config/
├── sources.json          # [BLOCKER] Data sources
└── keys.json             # API keys (gitignored)

scripts/
├── run_pipeline.py       # Main pipeline script
├── manage_actions.py     # Action tracker
└── collectors/           # Data collectors

references/
├── analysis-result-schema.md  # Detailed JSON schema
└── notion-page-template.md    # Report template

---

Notion Sync Flow (From MD File)

                    ┌──────────────────────────────────┐
                    │  Stage 5: MD Report Generation   │
                    └───────────────┬──────────────────┘
                                    │
                                    ▼
          ┌─────────────────────────────────────────────────┐
          │  reports/analysis-report-{date}.md              │
          │  ─────────────────────────────────              │
          │  • Executive Summary                            │
          │  • Bayesian Scores (tables)                     │
          │  • Pain Points (full evidence quotes)           │
          │  • Strategic Analysis (SWOT, competitors)       │
          │  • Action Items                                 │
          └───────────────────┬─────────────────────────────┘
                              │
                              │ READ (Only data source)
                              │
                              ▼
          ┌─────────────────────────────────────────────────┐
          │  Stage 6: Notion Sync                           │
          │  ─────────────────────                          │
          │  • Parse MD file structure                      │
          │  • Convert to Notion block format               │
          │  • Preserve content integrity (no simplification)│
          └───────────────────┬─────────────────────────────┘
                              │
                              ▼
          ┌─────────────────────────────────────────────────┐
          │  Notion Page                                    │
          │  ──────────                                     │
          │  Content == MD File Content ✓                   │
          └─────────────────────────────────────────────────┘

⚠️ Consistency Guarantee:

1. MD file is the Single Source of Truth (SSOT)
2. Notion sync MUST read from MD file
3. Three outputs must be consistent: MD file == Conversation output == Notion page