skills.rest
manhua-man
manhua-manCOMMUNITY
Explore all skills by this author →
GitHub

claude-session-distill

Distill Claude Code JSONL session transcripts into packets, notes, and verified knowledge.

1|Updated Jun 10, 2026
Search Tags:#claude code#knowledge base#batch processing#transcript analysis#jsonl#claim verification#session distillation
One-click install
npx skills add https://github.com/manhua-man/session-distill-skills --skill claude-session-distill-manhua-man
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: claude-session-distill
Source: https://github.com/manhua-man/session-distill-skills/tree/main/adapters/claude-session-distill
Command: npx skills add https://github.com/manhua-man/session-distill-skills --skill claude-session-distill-manhua-man

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Raw Claude Code session transcripts (.jsonl) accumulate as noisy, unverified chat logs. This Skill converts them into reviewable packets, structured session notes, and a canonical knowledge base, with a mandatory toolchain verification gate so only evidence-backed claims are promoted. ## Core Features & Use Cases - Packet Generation: Index and bundle Claude Code .jsonl sessions into lossless review packets with a Packet Audit exposing coverage and clipping. - Deep Distill Pipeline: Batch-process sessions (default 3 per batch) into answer-packets with question tables, then verify each claim via Grep/Read/git/Shell before promotion. - Knowledge Promotion Gate: Only rows marked ANSWERED are written to the canonical session-knowledge-base.md, with staleness audits marking outdated facts STALE or CONTRADICTED. - Use Case: After a week of Claude Code debugging sessions, run the batch distiller to extract reusable troubleshooting rules, verify them against the current repo HEAD, and promote only confirmed facts into the project knowledge base. ## Quick Start Ask the AI to run the session-distill deep distill batch on the next three pending Claude sessions and verify each extracted claim before promoting anything to the knowledge base.

Frequently Asked Questions about claude-session-distill

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

FAQPage Schema
How do I distill Claude Code session transcripts into reusable knowledge?▼

Run the session-distill CLI to index and bundle .jsonl sessions into packets, then write a session note per packet and append only stable, verified rules to the knowledge base. Finish by marking the session distilled so it leaves the pending queue.

What is the deep distill workflow for session transcripts?▼

Deep distill processes sessions in batches of three: generate packets, extract claims into answer-packets, verify each question with toolchain evidence, promote only ANSWERED rows to the knowledge base, and produce a check-work report before closing the batch.

When should I review the raw JSONL transcript instead of the packet?▼

Open the raw transcript whenever the Packet Audit shows Coverage: partial, when packet text appears clipped, or when a claim is a high-risk cross-module rule. The packet is the default entry point; raw files are the fallback evidence source.

Can session-distill sync extracted knowledge to claude-mem automatically?▼

No, claude-mem sync is an optional enhancement path, not the default chain. The packet-memory-export tool can generate draft memory entries labeled new, refine, confirm, conflict, or ephemeral, but human review decides what actually syncs.

Why must claims be verified before promotion to the knowledge base?▼

Transcript content is treated as hypotheses, not evidence. Each claim must be confirmed against the current repo using Read, Grep, git, or Shell, and only rows with status ANSWERED may be promoted; contradicted or stale claims are marked accordingly.

Related Skills

Explore More Software Engineering Skills→
Largo2z9Largo2z9
1

export-session

Export Claude Code sessions into structured markdown transcripts for archiving.

Community
Intermediate
SignalPilot-LabsSignalPilot-Labs
475

dbt-knowledgebase

Propose structured dbt knowledge entries across six categories via propose_knowledge.

Official
Advanced
flonatflonat
127

knowledge-lint

Automate knowledge base reviews for contradictions, uncited claims, and coverage gaps.

Community
Advanced
saxenautssaxenauts
58

syke

Maintain a local timeline of AI activity across supported platforms.

Community
Advanced