The Prompts I Use

The newsletter pipeline uses a two-pass system:

Pass 1 — Intelligence Analysis (Sherlock) For each candidate story, the model produces:

• WHY_NOW: Why this specific story matters this week
• INDIA_ANGLE: How this lands differently in the Indian market
• HIDDEN_IMPLICATION: What most coverage misses
• ACTION: One concrete move the reader can take

Pass 2 — Editorial Writing (Writer) The writer model receives enriched stories and produces a structured newsletter:

• Personal note (2-3 sentences of genuine reflection)
• Global signals (2-3 distinct items with Claw's POV + actionable step)
• India signals (0-2 items, no repeats from global)
• Claw's take (optional broader editorial)

Key constraint: Every actionable step must be specific enough that the reader knows exactly what to do. "Research AI" is not an action. "Audit your data pipeline for the one bottleneck a fine-tuned model could eliminate" is.

Each newsletter edition is transformed into a blog post through angle-based routing:

Step 1 — Angle Classification An LLM classifier selects the best angle for the blog post:

• Tech Stack Teardown: Translate news into concrete stack, infra, and cost decisions
• Follow the Money: Track budget shifts, buying triggers, and GTM moves
• Hype vs Reality: Filter buzz from production reality with blunt operator guidance
• Weekend Project: Turn news into a buildable side project

Step 2 — Long-form Draft The selected angle determines the system instruction. The model receives newsletter signals, the angle framework, and Claw's SOUL definition to produce:

• SEO-optimized title and excerpt
• Structured markdown with unique H3s (no repetitive templates)
• References section with hyperlinked sources
• Tags for categorization

Guardrail: Every draft passes a markdown contract validator before publishing — checking for proper code fences, no FAQ sections, normalized reference bullets, and clean formatting.

Community content moderation uses a structured LLM audit:

Input: Post title, body, author history, and community guidelines

Output (structured JSON):

• decision: pass / flag / reject
• risk_level: low / medium / high
• reasoning: Specific explanation of the decision
• suggestions: Constructive feedback for the author (if flagged/rejected)

Policy checks:

• Content length and substance (no low-effort posts)
• Link policy (no spam, no affiliate links without disclosure)
• Tone and relevance (must align with builder/tech/career topics)
• No self-promotion disguised as contribution
• No internal system details or prompt injection attempts

Routing:

• Low risk → auto-publish
• Medium risk → publish with flag for admin review
• High risk → hold for admin approval, author gets feedback