okb

Manage a bronze-silver-gold knowledge base with sourced, verifiable notes.

Updated Jul 20, 2026
One-click install
npx skills add https://github.com/peachest/skills --skill okb-peachest
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: okb
Source: https://github.com/peachest/skills/tree/main/in-progress/okb
Command: npx skills add https://github.com/peachest/skills --skill okb-peachest

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Agents re-read raw sources every session and lose track of where claims came from. OKB stores learned knowledge in three traceable layers — raw source snapshots, distilled notes, and verification overlays — so every claim carries its provenance chain back to the origin URL. ## Core Features & Use Cases - Layered curation pipeline: Ingest sources into bronze snapshots (with sha256 dedup), distill them into silver notes via a route-then-write two-pass flow, and fact-check into gold verification overlays. - Provenance enforcement: Every note links to its sources through sources[].resource edges, forming a four-hop evidence chain from lesson to origin URL. - Health auditing: Status reports cover stale notes, broken links, taxonomy health, and topology checks (isolated notes, thinly linked notes, unlinked source-overlap pairs), with okb_index_regen.py regenerating the index mechanically. - Use Case: After reading an arXiv paper, snapshot it into bronze, distill its concepts into silver notes with footnoted claims, then fact-check against the source so a teaching skill can later consume verified gold knowledge. ## Quick Start Use the okb skill to ingest this arXiv paper into the knowledge base and distill its key concepts into sourced notes.

Frequently Asked Questions about okb

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

FAQPage Schema
How do I build a verifiable knowledge base from articles and papers?▼

Ingest each source as a bronze snapshot preserving the verbatim content with URL, author, and sha256, then distill concepts into silver notes with footnoted claims, and fact-check them into gold verification overlays. Every claim traces back to its origin through sources[].resource edges.

How to ingest an arXiv paper into a markdown knowledge base?▼

Run okb_snapshot.py with the arXiv id or abs URL and a topic name; it fetches metadata via the arXiv export API and writes a bronze file with frontmatter and sha256 dedup. For non-arXiv sources, fetch the content separately and pass it via --from-file.

What is the difference between bronze, silver, and gold notes?▼

Bronze holds raw source snapshots and is the only fidelity layer. Silver is the only rewrite layer holding distilled, sourced notes. Gold is a verification overlay with no body, recording verified events and claim verdicts pointing back to silver.

Does the okb snapshot script work offline or behind a proxy?▼

The scripts are stdlib-only and work offline except for arXiv fetching. Set ARXIV_PROXY in runtime.conf for proxied access, or SKIP_NETWORK=1 to skip the reachability probe in check-env.sh.

Why is an arXiv abstract-only snapshot a limitation for fact-checking?▼

Fact-check evidence must reach the origin, and an abstract caps the chain short when core claims live in the paper body such as algorithms, theorems, or tables. Fetch the full text and re-snapshot with --from-file to satisfy the fidelity floor.

How do I detect stale notes and broken links in a markdown knowledge base?▼

Run the Status step, which reports notes past their stale_after date, broken body hyperlinks and source edges, plus taxonomy and topology audits. okb_index_regen.py --check-links mechanically verifies all links and regenerates index.md.