analysis-agent

The batch analysis pipeline (pipeline/analyze.ts) produces static output. With 19+ extracted calls, there's a need to ask ad-hoc questions — segment comparisons, pattern identification, positioning validation — in a conversational way, grounded in actual data with citations.

A /analyze Claude Code skill that loads call data and strategic goals, then answers questions with evidence: specific call references, verbatim quotes, quantitative breakdowns, and PMF-framed interpretation.

• Created goals.md — Static strategic analysis goals file (PMF questions, ICP signals, "what good looks like"). Loaded as context by the analysis agent.
• Created .claude/skills/analyze/SKILL.md — Interactive analysis agent skill. Read-only (Read, Glob, Grep). Loads goals.md + data/all-calls.json. Can read raw transcripts from data/raw/ for deeper context.
• Updated README.md — Documented the analysis agent, usage examples, and configuration.

Working /analyze skill tested with Cogent Security interest assessment. Agent loaded structured data, cited specific calls by company/date, included verbatim quotes, flagged urgency/budget gaps, and connected findings to PMF implications. Raw transcript access available for follow-up questions requiring exact wording.