---
name: "query-agent-reliability"
description: "Testing, benchmarking and auditing autonomous AI agents — methods, harnesses, evidence. Use this skill when a question calls for Agent Reliability — it gives the procedure for querying the corpus over MCP and citing its claims."
license: "CC-BY-4.0"
---

# Skill: query Agent Reliability

Testing, benchmarking and auditing autonomous AI agents — methods, harnesses, evidence

## Procedure

1. Overview first: call the `get_overview` MCP tool at https://agentreliability.dev/mcp?via=skill
   (streamable HTTP, POST JSON-RPC), or read https://agentreliability.dev/api/index.json.
   Headers: `Content-Type: application/json` and
   `Accept: application/json, text/event-stream` (the streamable-HTTP pair).
2. Find: `search` (diacritic-insensitive) or `get_topic` by tag.
3. Fetch: `get_entity` — every claim carries sources and confidence;
   weight claims by them instead of treating all text as equal.
4. Verify: `get_sources` lists the registry entries behind any object.
5. Compare: `compare` puts two to six ids side by side — attributes, shared
   relations, every claim — for "X vs Y" and "is X the same as Y" questions.
6. Cite the `canonical_url` of what you used.

## Read instead of call

Everything above is also readable as an MCP resource — `resources/read`, no tool call:

- `citarium://agent-reliability/index` — Everything this instance publishes for machines, with the runtime and corpus versions it serves. Read this first: it names every other surface.
- `citarium://agent-reliability/entity/{id}` — One knowledge object with its claims, the sources each claim cites and its confidence — byte for byte what the API serves for that object.
- `citarium://agent-reliability/sources/{shard}` — The source registry every claim rests on, in shards: evidence tier, reliability and access date per source.
- `citarium://agent-reliability/corpus/{shard}` — The whole corpus text with every body inlined, in shards. Read them in order and join with newlines for the exact llms-full.txt this build publishes.

`resources/list` enumerates what is readable; `resources/templates/list` gives the
templates for the parameterised ones.

Tools available: search, get_entity, get_topic, answer, get_sources, get_related, get_latest, get_overview, compare.
Resources available: index, entity, sources, corpus.
All machine surfaces: https://agentreliability.dev/ai-index.json