Glossary

Skill HelloThisWorld/agent-skill-verification-template/skills/glossary

Production-oriented template for building AI agent skills as verifiable software components — offline eval harness, source-grounding validators, structured logs/traces/metrics, replay artifacts, and a CI quality gate. Runs fully offline with a deterministic mock model.

Install

npx -y skills add HelloThisWorld/agent-skill-verification-template --skill glossary

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Looks up "glossary <term>" on English Wikipedia and returns a source-grounded definition rendered as a web page. Every claim cites the exact snapshot line carrying the term; unknown terms return insufficient_evidence instead of a guess.

SKILL.md

3.4 KB, as published. Nobody here has run it

Glossary (Wikipedia)

A Claude-style skill that answers glossary <term> requests by looking the term up on English Wikipedia and returning a definition rendered as a web page, where every factual claim is backed by a file:line citation into the fetched article. It is built to be verified like a production component — see skill-contract.json for the machine-readable contract and verification-rules.md for how outputs are graded.

When to use

Use this skill for requests such as glossary Mexico or glossary Switzerland. The term after glossary is looked up; the skill produces a concise, source-grounded definition and a self-contained HTML page for it.

Tools

Tool	Purpose
`wikipedia_search`	Search the offline Wikipedia snapshot cache. Returns `{title, file, line, text}` matches, best article first.
`wikipedia_fetch`	Read a snapshot and return its structured article data plus the citable "lede" line.

Contract rule: wikipedia_search must be used before wikipedia_fetch.

Offline-first design

Like the rest of this template, the default eval runs fully offline and deterministically. npm run glossary:build-cache fetches each term once from English Wikipedia (MediaWiki action API) and writes a citable snapshot to fixtures/wikipedia/<term>.html. Each snapshot embeds a machine-readable glossary-data JSON block and a lede line that contains the exact query term verbatim, so citations stay valid even when Wikipedia's canonical title or opening phrasing differs from the query. After the cache exists, no network is required to run or verify the skill.

Procedure

Parse the term from the glossary <term> request.
Use wikipedia_search to locate the best-matching snapshot.
Use wikipedia_fetch to read the snapshot and confirm the citable line.
Produce a structured answer whose claim cites that file:line.
If no snapshot matches the term, return insufficient_evidence with an empty claims array. Never invent a definition or a citation.

Output contract

The skill returns JSON with:

status: answered | insufficient_evidence | refused
answer: a short natural-language definition
claims: array of { text, citations: [{ file, line }] }
toolCalls: array of { tool, arguments }
confidence (optional): low | medium | high

The web-page deliverable (reports/latest/glossary/<term>.html plus an index.html) is rendered from the same grounded snapshot by src/skills/glossary/render.ts. See examples.md for a concrete input/output pair.

Design note: contract vs. model

This SKILL.md and the contract are model-independent — they describe what a correct answer looks like. How reliably a given model satisfies the contract is measured separately by the eval harness. The offline glossary adapter is a reference implementation that satisfies the contract deterministically; the glossary-flaky adapter perturbs it to demonstrate failure detection.