distill-knowledge

Curates durable task residue into canonical tracked AGENTS.md files or an explicit no-op.

2|Updated Apr 29, 2026
One-click install
npx skills add https://github.com/siarhei-belavus/agent-public --skill distill-knowledge-siarhei-belavus
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: distill-knowledge
Source: https://github.com/siarhei-belavus/agent-public/tree/main/skills/atelier-workflow/distill-knowledge
Command: npx skills add https://github.com/siarhei-belavus/agent-public --skill distill-knowledge-siarhei-belavus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? After a task is implemented and reviewed, valuable decisions, constraints, and gotchas risk being lost or dumped into noisy notes. This Skill distills a completed task packet into durable, canonical knowledge in tracked AGENTS.md files, or explicitly decides nothing is worth keeping. ## Core Features & Use Cases - Candidate triage: Promotes only residue that is durable, costly to forget, and not recoverable from code or tests, preferring no-op over weak prose. - Landing-zone selection: Places knowledge in the nearest stable tracked AGENTS.md owner boundary with the right section (Scope, Commands & Validation, Constraints, Decisions & Gotchas). - Governance and approval rules: Applies trivial hygiene fixes directly but requires explicit human approval for semantic changes, and never writes AGENTS.override.md. - Use Case: After finishing a multi-phase feature with code review, run this Skill in a fresh context to decide which non-obvious decisions belong in the module's AGENTS.md, prune stale guidance, commit the change, and report the outcome to the lead. ## Quick Start Run the distill-knowledge skill in a fresh context on the completed task packet to curate durable residue into the tracked AGENTS.md files or return an explicit no-op.

Frequently Asked Questions about distill-knowledge

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

FAQPage Schema
How do I distill task outcomes into AGENTS.md files?▼

Run this Skill in a fresh context after implementation and both review phases are clean. It reads the task packet, final code, tests, and diff, triages durable candidates, places them in the correct tracked AGENTS.md section, and commits before reporting the outcome.

What knowledge should be promoted into AGENTS.md documentation?▼

Promote only residue that is not recoverable from code or tests, durable beyond the current task, costly to forget, and not better encoded in code, names, or tests. Weak prose should be rejected in favor of an explicit no-op.

When should distillation result in a no-op?▼

Choose no-op when candidates are already obvious from code or tests, lack durability, or would turn AGENTS.md into a worklog. Mid-initiative passes default to no-op unless the packet's Next durable distill field says now.

Can this Skill create new AGENTS.md shards or edit AGENTS.override.md?▼

No. Creating a new tracked shard requires explicit human approval, and semantic additions, supersessions, or deletions also need approval. The Skill must never write AGENTS.override.md, which is treated as non-canonical.

Why must the distill phase run in a fresh context window?▼

A fresh context prevents implementation and review backstory from biasing curation decisions. The distiller should rely only on the packet, final code, tests, and diff so that only verified durable residue is promoted.