Name: process-distillation
Author: kimmywork

Imported from kimmywork/zlf (.agents/skills/process-distillation/SKILL.md). Install upstream with npx skills add kimmywork/zlf --skill process-distillation. Copyright stays with the author (MIT).

Process Distillation

Process

Gather context: phase executed, artifacts produced, review feedback report, fix outcomes, execution observations. If context insufficient, explore agent session logs or memory stores as fallback.

Analyze gaps:

Dimension	Question
Coverage gap	Step skill should describe but didn't?
False guidance	Skill misled executor?
Missing guardrail	Review caught something skill should prevent?
Repeated improvisation	Executor re-invented same helper?
Atomic extraction	Clear bounded sub-process worth its own skill?
Context overhead	Skill too large, wasting context?
Trigger misfire	Description caused wrong trigger behavior?

Evaluate options: small fix / new reference / new atomic skill / no change. Prefer small fix (1–5 lines) over new skill. New skill creation always requires user approval.
Self-review: every claim must cite file evidence. Before rename, inventory affected files with grep. After changes, re-grep to confirm zero remaining hits. Use references/rename-checklist.md for cross-family renames. Classify remaining hits: intentional / false positive / needs fix.
Apply under AGENTS.md constraints: agent neutral, size controlled, scope limited, atomic, user-centric, English. SKILL.md is for the agent: precise instructions, no verbose rationale. When creating new skills, pass these constraints into the new skill's instructions.
Output: distillation report using references/distillation-template.md.
Approval: default is user approval. full-autonomy mode: auto-approve safe improvements. New skill creation always requires user approval.

Subagents

Available: use reviewer for gap analysis, writer for drafting. Separate analysis from authoring. Unavailable: self-perform, record limitation.

Typically follows review-feedback. In solution-delivery-loop, auto-triggers after resolved review-feedback cycles under full-autonomy.

{ "ocm": "1", "id": "kimmywork-zlf-process-distillation", "kind": "skill", "name": "process-distillation", "description": "Use when recurring phase friction, repeated review feedback, or resolved phase cycles suggest skill or process improvements.", "publisher": "kimmywork", "version": "1.0.0", "capabilities": { "domains": [ "customer_support", "observability" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Use when recurring phase friction, repeated review feedback, or resolved phase cycles suggest skill or process improvements." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/kimmywork/zlf", "path": ".agents/skills/process-distillation/SKILL.md", "ref": "fc4612efc00adc946e90f11673db0301e61ac135", "url": "https://github.com/kimmywork/zlf/blob/fc4612efc00adc946e90f11673db0301e61ac135/.agents/skills/process-distillation/SKILL.md", "key": "kimmywork/zlf/.agents/skills/process-distillation/SKILL.md" }, "license": "MIT" }, "instructions": "# Process Distillation\n\n## Process\n\n1. **Gather context**: phase executed, artifacts produced, review feedback report, fix outcomes, execution observations. If context insufficient, explore agent session logs or memory stores as fallback.\n\n2. **Analyze gaps**:\n\n | Dimension | Question |\n |---|---|\n | Coverage gap | Step skill should describe but didn't? |\n | False guidance | Skill misled executor? |\n | Missing guardrail | Review caught something skill should prevent? |\n | Repeated improvisation | Executor re-invented same helper? |\n | Atomic extraction | Clear bounded sub-process", "cost": { "context_tokens": 510 } }

process-distillation

About