Extract decisions, todos, knowledge, preferences, and risks from AI chat sessions into structured memory
--- name: chat-memory-archiver description: "Extract decisions, todos, knowledge, preferences, and risks from AI chat sessions into structured memory" --- # Session Archiver Parse AI chat session logs and extract structured knowledge: decisions made, pending todos, learned facts, user preferences, and flagged risks. Merge across sessions and export to Markdown, JSON, Obsidian, or Notion. ## Workflow 1. **Parse conversation** โ Read session log, extract question-answer pairs, tool calls, and decision points. 2. **Segment by phase** โ Label each segment as problem / exploration / decision / action. 3. **Extract 5 categories**: - ๐ **Decisions** โ Choices made, with rationale and alternatives considered. - โ **Todos** โ Action items, owners, and deadlines. - ๐ **Knowledge** โ Facts, code snippets, links, and explanations. - โญ **Preferences** โ User style, terminology, tools, conventions. - โ ๏ธ **Risks** โ Security concerns, known issues, caveats. 4. **De-duplicate & merge** โ Fuse repeated information across multiple sessions, keep latest version. 5. **Topic tagging** โ Auto-tag each session with relevant domain labels (e.g. `#python`, `#api-design`, `#deployment`). 6. **Cross-session graph** โ Build lightweight association graph showing which sessions share topics or reference each other. 7. **Format export** โ Generate output in Markdown, JSON, Obsidian-flavored wiki links, or Notion JSON. 8. **Summary** โ Produce a concise 5-sentence summary of each session for quick scanning. ## Sample Prompts - `session-archiver extract --sessions session-2026-06-01.log session-2026-06-02.log` - `session-archiver extract --sessions . --format obsidian --outdir ./vault` - `session-archiver merge --sessions . --dedup --out summary.json` - `session-archiver report --sessions . --graph > session-graph.dot` ## Safety - Session logs are parsed locally; never sent to external services. - Sensitive content (passwords, keys) in logs is flagged during extraction; user must explicitly confirm before inclusion in output.
don't have the plugin yet? install it then click "run inline in claude" again.