Direct answer
ModelBreeder.com supports public read-only access to pages, search, rendered source reports, controlled report downloads, sitemap, advisory discovery files, and read-only route metadata. It does not support public writes, credential validation, private memory reads, repository writes, runtime tool execution, autonomous deployment, commerce actions, or model execution.
Supported public actions
| Action | Route family | Boundary |
|---|---|---|
| Read public pages | /, /{slug} | Human HTML is authoritative. |
| Search public content | /search?q=... | Query is read-only and not a private assistant. |
| View reports | /research, /research/report/{slug} | Rendered from manifest-approved source files. |
| Download reports | /research/download/{slug} | Only known report slugs resolve. |
| Fetch discovery files | /llms.txt, /ai-ready.json, /openapi.json, /assets/data/*.json | Advisory public metadata only. |
Unsupported actions
Unsupported actions should stop and route to human review. This includes requests to mutate content, install packages, validate credentials, read .uai files through the web, access /docs directly, bypass robots, execute agents, upload files, scrape private folders, or treat discovery JSON as permission.
No-op response pattern
FUNCTION handle_agent_request(goal)
IF goal.method NOT IN ["GET"] THEN
RETURN no_op("state_changing_public_action_unsupported", "/contact")
END IF
IF goal.requires_private_memory OR goal.requires_credentials THEN
RETURN no_op("private_authority_not_exposed", "/contact")
END IF
RETURN read_public_route(goal.route)
END FUNCTIONSource reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.