Documentation library

Written for agents first. Every page here is also JSON and markdown from the same URL: send Accept: application/json or Accept: text/markdown. Reads need no token.

For an agent: read this index as JSON, pick a use case, fetch its OpenSpec plan (/use-cases/{slug}/openspec?lang=http|python|typescript&with=commitops,repoops,promptreports), write the files into openspec/changes/ in your repo and implement them.

Start here

• Agents: register, get a token, publish, verify, MCP tools, OAuth
• llms.txt: the whole surface on one page
• For AI assistants: how to cite a record in an answer
• Publish: the publish contract and install lines that work today

Interfaces

• MCP server: Streamable HTTP, 37 tools, read tools need no token
• OpenAPI: every HTTP route
• Record schemas: JSON Schemas for every record profile, CC0
• Agent card: A2A agent card
• SCITT: submit COSE signed statements, get COSE receipts

Use case plans

• All use cases: 13 workflows, grouped
• Cite a source and recheck it later: OpenSpec plan
• Filter what you crawl or train on: OpenSpec plan
• Check an AI disclosure: OpenSpec plan
• Verify before an agent pays: OpenSpec plan
• Attribute published AI content: OpenSpec plan
• Build on another publisher's work: OpenSpec plan
• AI-written code, tied to the pull request: OpenSpec plan
• Verified marketplace goods: OpenSpec plan
• Proof that a work existed by a date: OpenSpec plan
• Regulated audit trails: OpenSpec plan
• Evidence an investigator can take to court: OpenSpec plan
• Trace an incident to the prompt: OpenSpec plan
• Creators paid by any platform: OpenSpec plan

Guides

• Verify before paying: ACP, UCP and x402: Where a buying agent calls verify-batch in each agent payment flow, what to do with a seal or an erratum, and how to witness a payment mandate and its receipt by digest.
• C2PA lookup for providers: Register C2PA manifest digests and soft binding ids so a stripped file resolves back to its record. The registry runs no detection. Not legal advice.
• Record fields and EU AI Act and NIST asks: A reference map from registry record fields to the record-keeping and transparency asks in the EU AI Act and the NIST AI RMF. Not legal advice.
• Anchor an agent audit log: Post the head of a halo-record, agentproofs or other hash-chained agent log to the registry on a timer, so a later rewrite of the log shows.

Check it yourself

• Assess: rely, check or do_not_rely, with reasons
• Verify batch: POST up to 1000 digests, no token
• Offline verifier: check an export bundle with no network
• Error contract: typed errors and how to fix each

Definitions

• What is a verification record: A signed, content-addressed record of what was verified, by whom, under which judge, when.
• What is a provenance record: The minimal, self-serve attribution record: who or what made an asset, with which model, from which inputs.
• What is a judge hash: The hash that pins the exact judge model and prompt version a verification record was produced under.
• How AI-created assets get a permanent public page: Every AI-created asset gets one permanent, canonical, citable page aggregating its records, lineage, and errata.
• Verify-before-transact for agent commerce: Any agent, transacting anywhere, can gate on a registry check before it spends.

With the PromptReports products

• CommitOps: keeps the ledger of what a company promised its customers, and can tie a promise to the record that shows it shipped
• RepoOps: records what coding agents did in a repo: sessions, agent versions and incidents
• PromptReports: publishes research reports, each with its own signed record