Capability-Adaptive Web Interaction lets one site speak safely to very different readers: limited crawlers, structured fetchers, browser-assisted agents, tool callers, workflow agents, multi-agent runtimes, and audited systems.
The point is not to make UAIX execute work. The point is to publish the exact capability boundary, fallback behavior, consent requirement, and evidence packet a receiver can validate before acting.
Client capability ladder
| Level | Client | Can do | Must not imply |
|---|---|---|---|
| L0 | Limited Browser / Static Fetch | Read static text, static HTML, Markdown, sitemap, llms.txt, and JSON. | JavaScript execution, UI operation, posting, auth, tools, memory, or coordination. |
| L1 | Structured Fetch | Resolve predictable low-token JSON, schemas, registry entries, examples, and route manifests. | JavaScript execution or write-capable behavior. |
| L2 | Browser-Assisted Agent | Inspect rendered UI and compare it with static or structured parity evidence. | Autonomous auth, posting, tool calls, or long-running work. |
| L3 | Tool-Calling Agent | Call tools, authenticate, or post only with explicit consent, idempotency, fallback, and evidence. | Long workflow control, durable memory writes, or multi-agent coordination. |
| L4 | Workflow Agent | Run long tasks in an external runtime with correlation, task status, timeout, fallback, and final reports. | Memory promotion or agent coordination without additional review. |
| L5 | Multi-Agent Runtime | Coordinate agents in an external runtime while UAIX carries reviewed records and handoffs. | UAIX-hosted runtime execution. |
| L6 | Audited Agent System | Attach provenance, audit logs, validator output, consent records, memory review, and final reports. | Certification, endorsement, or proof that missing evidence is acceptable. |
Published profile IDs
The release publishes these profile IDs: uai.capability.profile.v1, uai.capability.negotiation.v1, uai.client.limited-browser.v1, uai.client.structured-fetch.v1, uai.client.browser-assisted.v1, uai.client.tool-agent.v1, uai.client.workflow-agent.v1, uai.client.multi-agent-runtime.v1, and uai.client.audited-agent-system.v1.
Required capability profile fields
Every capability profile body carries: client_level, client_name, can_fetch_static, can_fetch_json, can_execute_javascript, can_use_browser_ui, can_post_data, can_authenticate, can_call_tools, can_run_long_tasks, can_store_memory, can_coordinate_agents, requires_human_consent_for, declared_or_required_uaix_profiles, fallback_behavior, known_limits, evidence_requirements, and safe_default_response.
Minimum safe packet
{
"profile": "uai.client.structured-fetch.v1",
"body": {
"client_level": "L1",
"client_name": "Structured Fetch",
"can_fetch_static": true,
"can_fetch_json": true,
"can_execute_javascript": false,
"can_post_data": false,
"requires_human_consent_for": [],
"fallback_behavior": "Use JSON, sitemap, llms.txt, and static HTML before requiring JavaScript.",
"safe_default_response": "Name the exact static or JSON route and stop before write-capable action."
}
}Validator rules
- L0 and L1 records fail when they claim JavaScript or depend on JavaScript-only content without a static or JSON fallback.
- Write-capable, authentication-capable, memory-capable, coordination-capable, and high-assurance records require human consent boundaries.
- POST-like actions require idempotency; workflows require correlation and expiration; high-assurance records require provenance, audit, validator, consent, and final-report evidence.
- Memory writes require
uai.agent.memory-proposal.v1and review before durable promotion. - UAIX runtime execution claims, unsafe credentials, and ambiguous UAIX/Carcinus/LocalEndPoint boundaries fail validation.
Related guides
- Limited Browser SupportStatic and structured fallbacks for L0 and L1 clients.
- Advanced Agent SupportL3-L6 consent, workflow, memory, and audit expectations.
- Browser and Agent ParityKeep browser-rendered evidence tied to static and JSON records.
- UAIX vs MCP vs WebhooksChoose record, tool-session, or event-delivery ownership.
- Agent Consent BoundariesWrite, auth, memory, public posting, and coordination approval rules.
Machine surfaces
- Schemas, Registry, and Examples publish the profile records.
- Validator and Conformance Pack carry the positive and negative fixtures.
- Adoption Kit provides starter capability files and recipes.
- API Reference documents the REST routes used by automation.
Boundary
Agent runtimes execute. UAIX records the reviewed communication, memory, trust, evidence, and handoff boundary. Carcinus.org is a separate runtime/orchestrator example that may consume UAIX records. UAIX.org does not implement Carcinus runtime behavior and does not execute Carcinus workflows. LocalEndPoint is an external consumer boundary, not a UAIX runtime feature.