Mines past Claude Code session transcripts to draft reusable skills from completed workflows.
npx skills add https://github.com/yang-l/configs --skill distill-skill-yang-l
Please help me install this Agent Skill. Skill: distill-skill Source: https://github.com/yang-l/configs/tree/main/.claude/skills/distill-skill Command: npx skills add https://github.com/yang-l/configs --skill distill-skill-yang-l
.jsonl session transcripts under ~/.claude/projects/, including subagent transcripts, using a stdlib-only Python helper script. - Structured extraction with safeguards: Delegates interpretation to a read-only researcher agent that produces what/command/verify/why steps, tags unsafe verification commands, redacts credentials and hostnames, and flags conflicts with feedback memories. - Human-gated authoring: Presents the draft for explicit approval, then hands off to skill-creator for authoring, validation, and gitignore-by-default writing. - Use Case: You fixed a tricky deployment issue last week and want it captured. Ask to turn that past session into a skill, and this Skill locates the transcript, extracts the verified procedure, and drafts it for your review. ## Quick Start Turn the workflow from my session last week where I fixed the deploy pipeline into a reusable skill.High-intent search queries and answers about installing and using this skill.
Reference the past work in plain language, such as naming the session or the task you completed. The Skill lists matching session transcripts, filters the relevant commands and decisions, drafts the procedure, and hands it to skill-creator after your approval.
The bundled mine_sessions.py script lists sessions by matching a project token against real directory names under ~/.claude/projects/, then builds a dated, titled manifest from ai-title and last-prompt records inside the .jsonl files. It never reads sessions-index.json, which is stale.
No. It only mines, filters, and drafts the procedure, then presents the draft for explicit human approval. The actual authoring and validation are delegated to skill-creator, and the new skill directory is appended to .gitignore by default.
The Skill stops and reports that no clear procedure was found rather than fabricating steps. The helper script exits non-zero with an explicit message on empty results, and the researcher agent flags partial or missing procedures as open questions.
The extraction step redacts account IDs, ARNs, tokens, and internal hostnames to placeholders before drafting. Combined with gitignore-by-default on write, this prevents sensitive content from being committed to a public repository.
Use skill-creator when you want a brand-new skill with no prior session to reference. Distillation only applies when the workflow was already performed in a past session whose transcripts can be mined.
duthahoMine Claude Code transcripts to rank repeated corrections and propose harness improvements.
JhonMA82Generates verifiable Comet-native workflow Skill bundles from confirmed user proposals.
agentuityExtracts structured conclusions from session data using formal reasoning types and validity checks.
lc0rpReview live Codex threads to identify skill gaps and recommend updates.