distill

Analyze merged PR diffs and issue STM evidence to write learning proposals.

3|Updated Jan 25, 2026

Search Tags:#agentic workflow #git diff #fire-and-forget #post-merge #learning extraction #fast mode #STM proposals

Ready to run

npx skills add https://github.com/kapilvirenahuja/garura --skill distill-kapilvirenahuja

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: distill
Source: https://github.com/kapilvirenahuja/garura/tree/main/core/components/plays/distill
Command: npx skills add https://github.com/kapilvirenahuja/garura --skill distill-kapilvirenahuja

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It turns merged PR diffs plus any available issue STM evidence into lightweight, staged learning proposals without blocking or halting the ship workflow.

Core Features & Use Cases

Post-merge learning extraction (FAST mode): Analyzes merged PR diffs and relevant STM evidence to produce 1–2 learning proposals.
STM-only staging: Writes outputs to issue STM at {stm_base}/{issue}/evidence/distill/ and avoids product LTM writes.
Fire-and-forget safety: Catches and logs failures gracefully so the calling ship play proceeds to scenario validation.
Trivial PR handling: Produces no proposals for trivial PRs, returning a completed result with no_learnings.

Quick Start

Run distill after merge-pr to stage learning proposals to issue STM at the distill evidence path when evidence.record is enabled.

Frequently Asked Questions about distill

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

FAQPage Schema

How do I extract post-merge learnings from a git diff into STM?▼

To extract post-merge learnings from a git diff into STM, analyze merged PR diffs alongside available issue STM evidence. This distillation process generates lightweight learning proposals and stages them safely without halting your workflow.

Can I automate learning extraction from merged PRs without blocking the ship workflow?▼

Yes, you can automate learning extraction from merged PRs without blocking the ship workflow using a fire-and-forget approach. It catches and logs any failures gracefully, ensuring the calling ship play proceeds to scenario validation without ship-halting exceptions.

How does post-merge learning distillation handle trivial PR diffs?▼

Post-merge learning distillation handles trivial PR diffs by safely skipping them. It analyzes the diff and produces no proposals for trivial changes, returning a completed result with no_learnings instead of staging unnecessary data.

Where are post-merge learning proposals staged in the issue STM directory?▼

Post-merge learning proposals are staged exclusively in the issue STM at the path {stm_base}/{issue}/evidence/distill/. The process writes a proposals.yaml file to this directory and intentionally avoids writing to product LTM.

Do I need FAST-mode invocation of knowledge-extractor to distill PR diffs?▼

Yes, you need FAST-mode invocation of knowledge-extractor to distill PR diffs. This mode is required to process the merged PR diffs and relevant STM evidence to produce 1–2 learning proposals efficiently.

What is the best way to stage ship-after-merge learning proposals for non-trivial PRs?▼

The best way to stage ship-after-merge learning proposals for non-trivial PRs is to run a distill operation after merge-pr. This fits ship-after-merge scenarios perfectly when evidence recording is enabled, generating staged proposals without halting.