analyze-project

/analyze-project — Root Cause Analyst Workflow

Analyze AI-assisted coding sessions in brain/ and produce a diagnostic report that explains not just what happened, but why it happened, who/what caused it, and what should change next time.

This workflow is not a simple metrics dashboard. It is a forensic analysis workflow for AI coding sessions.

---

Primary Objective

For each session, determine:

1. What changed from the initial ask to the final executed work
2. Whether the change was caused primarily by:
   • the user/spec
   • the agent
   • the codebase/repo
   • testing/verification
   • legitimate task complexity
3. Whether the original prompt was sufficient for the actual job
4. Which subsystems or files repeatedly correlate with struggle
5. What concrete changes would most improve future sessions

---

Core Principles

• Treat .resolved.N counts as signals of iteration intensity, not proof of failure
• Do not label struggle based on counts alone; classify the shape of rework
• Separate human-added scope from necessary discovered scope
• Separate agent error from repo friction
• Every diagnosis must include evidence
• Every recommendation must map to a specific observed pattern
• Use confidence levels:
  • High = directly supported by artifact contents or timestamps
  • Medium = supported by multiple indirect signals
  • Low = plausible inference, not directly proven

---

Step 1: Discovery — Find Relevant Conversations

1. Read the conversation summaries available in the system context.
2. List all subdirectories in: `~/.gemini/antigravity/brain/
3. Build a Conversation Index by cross-referencing summaries with UUID folders.
4. Record for each conversation:
   • conversation_id
   • title
   • objective
   • created
   • last_modified
5. If the user supplied a keyword/path, filter on that. Otherwise analyze all workspace conversations.

Output: indexed list of conversations to analyze.