skills.rest
nickgogan
nickgoganCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

source-triage

Scans research sources and assigns extract, skip, defer, or link-only verdicts.

Updated Apr 19, 2026
Search Tags:#research#knowledge base#subagent#deduplication#extraction#verdict#source triage
One-click install
npx skills add https://github.com/nickgogan/improvement-loop --skill source-triage-nickgogan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: source-triage
Source: https://github.com/nickgogan/improvement-loop/tree/main/.claude/skills/source-triage
Command: npx skills add https://github.com/nickgogan/improvement-loop --skill source-triage-nickgogan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Deciding which research sources deserve full extraction is slow and error-prone when done manually, especially after a source quality audit flags dozens of unlinked sources. This Skill quickly estimates finding density per source and produces a prioritized triage report so extraction effort goes only where it pays off. ## Core Features & Use Cases - Verdict Assignment: Applies a decision tree to label each source EXTRACT, SKIP, DEFER, or LINK-ONLY based on actionable pattern counts and existing knowledge base coverage. - Parallel Subagent Batching: Groups sources into batches of 5-8 and dispatches subagents that skim content via WebFetch and return strict JSON verdicts. - KB Deduplication: Builds a lightweight index of existing findings to avoid re-extracting patterns the knowledge base already covers. - Use Case: After a source quality audit identifies 40 sources with zero linked findings, run the triage to discover that 12 warrant full extraction, 8 only need linkage repair, and 20 can be skipped, then queue each group for the right follow-up workflow. ## Quick Start Run source-triage with --all-unlinked to triage every source that has no linked findings and generate a prioritized report.

Frequently Asked Questions about source-triage

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

FAQPage Schema
How do I decide which research sources are worth extracting findings from?▼

Run source-triage on the candidate sources to get EXTRACT, SKIP, DEFER, or LINK-ONLY verdicts. It skims each source, counts actionable patterns, checks them against existing knowledge base findings, and outputs a prioritized report sorted by novel pattern count.

How do I triage sources with no linked findings after a quality audit?▼

Use the --all-unlinked flag to automatically select all sources whose findings list is empty, or use --tier to target a specific tier from the source quality audit report. The skill batches them and processes batches in parallel subagents.

Does source-triage create findings or update source metadata?▼

No, it never creates findings; that is the research-loop skill's job. It only adds an optional triage note to the source body recording the verdict and reason, leaving frontmatter and source status unchanged.

Can source-triage evaluate video sources without transcripts?▼

Yes, but with lower confidence. It relies on the source's key_takeaways, title, and tags, and notes the limitation in the verdict reason. The verdict may change once a transcript becomes available.

What is the difference between SKIP and LINK-ONLY verdicts?▼

SKIP means the source has fewer than two novel actionable patterns and needs no further action. LINK-ONLY means the content is already captured by existing findings but the source file is not linked to them, so it should be queued for linkage repair instead of extraction.

When should I not use source-triage?▼

Avoid it when you already know a source is high-value and needs full extraction, since triage only skims content for 30-60 seconds per source. It is a screening step, not a replacement for deep extraction or transcript fetching.

Related Skills

Explore More Education & Research Skills→
onebrain-aionebrain-ai
23

research

Research arbitrary topics from scratch and save structured Markdown notes.

Official
Intermediate
kiurakkukiurakku
15

natural-review

Diagnose AI writing tells in prose and produce a defect list with quoted evidence.

Community
Advanced
JansenAnalyticsJansenAnalytics
5

source-scorer

Score web source credibility by domain tier, content signals, and claim support.

Community
Intermediate
MykhailoDmytriakhaMykhailoDmytriakha

32-analyze-verify-150

Verify factual claims with primary sources, confidence scores, and Loop150 escalation.

Community
Advanced