logan CLAUDE.md

LOGAN — Log Analyzer

LOGAN is an Electron-based log analysis tool with an integrated AI agent interface.

Agent Communication

You can communicate with the LOGAN user through its built-in Chat panel. The MCP server is configured in .mcp.json at the project root.

Chat Loop Pattern

Use this pattern for interactive conversations with the LOGAN user:

1. logan_send_message("Hello! How can I help?")
2. response = logan_wait_for_message(timeout=120)
3. Process response, reply with logan_send_message(...)
4. Repeat from step 2 until user says goodbye

Session Memory — persist context across reconnects

LOGAN gives each open log a small per-file agent-memory scratchpad (stored at .logan/<file>.agent-memory.json, surfaced in the Chat panel's memory bar). It is NOT filled automatically — it stays empty until you write to it. Use it so you can resume naturally if the session drops and reconnects.

• At the start of a session (and after any reconnect), call logan_memory_read to recall what a prior session established.
• After each significant finding or task, call logan_memory_write(content=...) with a brief note of what the user asked and what you found. Each write replaces the previous note — write the full running summary, not just a delta.
• Memory is per-file: it's keyed to the currently open log, so a note written while one file is open won't appear when a different file is open. Writing with no file open is a no-op.

If the memory bar "keeps empty," it's almost always because no agent ever called logan_memory_write — the plumbing round-trips fine, it just isn't automatic.

Surfacing Findings — CRITICAL RULE

Whenever you identify a specific critical point, anomaly, or root cause in the log, you MUST call logan_report_finding to pin it in the viewer. Do NOT just describe findings in chat text — call logan_report_finding first, then follow up with explanation. This creates a visible annotation in the log viewer so the user can click to navigate directly to the issue.