A public knowledge contract for agents should expose compact and full content layers, freshness, confidence, provenance, intended use, non-use boundaries, search, versions, and stable discovery paths. Anything less leaves too much to guesswork.
Public Knowledge Contracts for Agents
If a page is meant to be read by agents, it should not make the agent reverse engineer the contract.
The contract should be public.
Minimum Page Fields
A serious page should expose:
- title
- summary
- compact summary
- updated date
- version
- confidence
- intended use
- do not use for
- sources
- change summary
Those fields let an agent answer basic questions about relevance, freshness, and safety without reading the whole essay first.
Layered Retrieval
At minimum, the site should offer:
metacompactfullraw
Compact is the first-pass route. Full is for drill-down. Raw is for exact source recovery. Meta exists so systems can inspect the frame before loading content.
Discovery And Routing
If the pages exist but agents cannot discover them, the contract is incomplete.
That is why a public agent surface should also provide:
- capabilities
- policy
- search
- discovery files
- version history
This turns the site from a passive set of documents into something agents can route through intentionally.
Boundaries Matter
Public knowledge is not the same thing as authorization.
Agents should treat the site as a trustworthy advisory surface, not as a hidden permission layer to execute actions in the world. That distinction is why trust metadata belongs inside the contract itself.