session-distill

Distills Cursor Composer session transcripts into verified repository knowledge base entries.

1|Updated Jun 10, 2026
One-click install
npx skills add https://github.com/manhua-man/session-distill-skills --skill session-distill-manhua-man
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: session-distill
Source: https://github.com/manhua-man/session-distill-skills/tree/main/adapters/cursor-session-distill
Command: npx skills add https://github.com/manhua-man/session-distill-skills --skill session-distill-manhua-man

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? AI coding sessions in Cursor accumulate as raw SQLite and JSONL transcripts full of unverified claims. This Skill distills those transcripts into structured packets, verifies each claim against the actual codebase with toolchain evidence, and promotes only confirmed facts into a canonical session knowledge base. ## Core Features & Use Cases - Batch Deep Distillation: Processes 3 Cursor sessions per batch via deep-distill-run.py, generating lossless packets and answer-packets with extracted claims. - Verification-Gated Promotion: Every claim must be verified with Grep/Read/git/Shell evidence and marked ANSWERED before promotion to the canonical knowledge base; contradicted or stale claims are flagged. - Storage Cleanup: Purges archived sessions from Cursor's state.vscdb, conversation-search.db, workspace storage, and agent-transcripts with automatic backups. - Use Case: After weeks of Cursor Composer sessions on a project, run a deep distill batch to extract reusable lessons, verify them against the current repo HEAD, and append only proven facts to session-knowledge-base.md. ## Quick Start Ask the AI to run the Cursor deep distill workflow on the next batch of three sessions and promote only toolchain-verified answers to the session knowledge base.

Frequently Asked Questions about session-distill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I distill Cursor Composer session history into a knowledge base?▼

Run deep-distill-run.py with a batch size of 3 to generate packets and answer-packets from Cursor sessions. Then verify each claim with toolchain checks like Grep or git, and promote only rows marked ANSWERED into session-knowledge-base.md.

How to extract chat history from Cursor state.vscdb SQLite database?▼

Cursor stores composer headers in ItemTable and conversation bubbles in cursorDiskKV within state.vscdb. The cursor-session-distill.py script reads these tables in read-only mode and falls back to JSONL agent-transcripts when SQLite bubbles are missing.

Can I delete archived Cursor sessions from the sidebar?▼

Yes, but you must clean both state.vscdb and conversation-search.db, since the sidebar Archived list is driven by the search database. Use purge-archived-cursor-sessions.py with Cursor fully closed; it creates backups automatically.

Why did my Cursor session cleanup not remove archived entries from the UI?▼

Cleaning only state.vscdb leaves conversation-search.db intact, and the sidebar reads archived entries from that search database. You must purge both databases and rebuild the FTS index for the UI to update.

What are the limitations of promoting session claims to a knowledge base?▼

Chat transcripts are treated as hypotheses, not evidence, so unverified claims cannot be promoted. Only rows verified against the current repo with toolchain proof and marked ANSWERED may enter the knowledge base; stale or contradicted claims stay in session notes.