UAIX guides turn the public records into practical adoption paths. Use these pages when the specification is not enough by itself and a team needs a concrete setup, operating sequence, or integration model.
Recommended reading paths
- UAIX / UAI-1 Agent Skill BundleDownload the reference skill bundle and use it to create, validate, cite, package, and hand off UAI-1 evidence.
- UAIX Talisman ClientDownload the WordPress client-node package for protected-anchor readout, hub sync, and request-only talk-back evidence.
- Agentic Harness Strategies And UAISeparate runtime harness strategy from the portable evidence, handoff, and conformance records UAI should preserve.
- Agentic systems and public evidenceUse Standards Fit to decide what the runtime owns, what MCP or A2A owns, and what UAIX should preserve as a portable public record.
- Using Project Handoff With OpenAI Agents And CodexUse OpenAI to run the agents while Project Handoff preserves the repo-local project memory.
- Long-Running Goal ExecutionRun interruptible goals with checkpoints, blockers, handoffs, final reports, and reviewed write-back.
- Hierarchical Goal MemoryFold completed subgoals into source-linked records and retrieve only bounded evidence for the next slice.
- Using Project Handoff With Coding AgentsUse one local handoff bundle across Codex, Claude Code, Cursor, Copilot, and Gemini Code Assist.
- Project Handoff Context BudgetPrevent AGENTS.md, .uai/readme.human, and .uai files from becoming oversized archives.
- Project HandoffCreate the durable AGENTS.md, .uai/readme.human, and .uai bundle.
- AI MemoryStart with the broader portable memory framing and starter bundles.
- RoadmapCheck current versus planned tooling before making support claims.
Agentic system evidence path
For agentic systems, start by separating the live control plane from the durable record. Runtimes, workflow engines, tool protocols, guardrails, and observability systems run the work. UAIX records the part that another team, release, auditor, customer, or future agent needs to validate and carry forward.
| Layer | Usually owns it | What UAIX should preserve |
|---|---|---|
| Execution and control flow | Agent runtime or workflow harness | Reviewed request, outcome, task status, handoff summary, and release evidence. |
| Tools and resources | MCP, API clients, local integrations, or runtime adapters | Portable request/result evidence when the interaction crosses a team, vendor, audit, or release boundary. |
| Agent coordination | A2A or runtime-native handoff | Capability statements, task-state records, and validator-ready public exchange packets. |
| Project memory | AI Memory and Project Handoff after human review | Current constraints, decisions, owners, evidence paths, checks, and receiver briefs that survive the run. |
| Trace and approval history | Runtime trace, approval, and observability systems | Only the reviewed, redacted identifiers and summaries needed to reproduce or cite the public record. |
Boundary: this path does not create SDKs, CLIs, official adapters, certification, endorsement, hosted import validation, automatic repository writes, or automatic LLM Wiki sync. Use the Roadmap before repeating those ideas as current support.
Agent Access
- Chatbot AccessPublic floor for URL-only chatbots and low-capability agents.
- Minimal Access TierRead-only GET, no body, no auth, no custom headers, two-field response.
- GET-Action PatternRequired for bounded idempotent GET-Action fallback configuration, separate from Minimal Access and not a POST replacement.
- Agent Capability LadderL0-L5 progressive web-access view.
- Progressive Agent AccessEscalate from public URL reads to POST JSON, manifests, and handoff evidence.
- GET-Action SecurityNo secrets in query strings, consent, rate limits, crawler protection, and audit rules.
Agent Compatibility
- Agent CompatibilityLow-capability quick start, GET-Action standardization, no-op behavior, and recursive interoperability checks.
- Capability-Adaptive InteroperabilityL0-L6 support contracts and fallback behavior.
- Agent Fallbacks and Full-Spectrum SupportRamp up to richer paths when proven and ramp down to safe fallbacks when unavailable.
- Agent Executability MatrixL0-L7 executable paths, route-record fields, result URLs, restore routes, and fallback obligations.
- GET-Action PatternLiteral semantic keys, URL-first fallback, POST equivalents, and code/url responses.
- GET-Action SecurityNo secrets in URLs, idempotency or nonce, consent, rate limits, and crawler protection.
- No-Op ProtocolUnsupported agents stop safely with a public review URL.
- Capability Surface MatrixExpected support by client level and surface type.
- Validator RulesRequired failures and warnings for agent compatibility records.
- Implementation ExamplesSafe GET-first and POST-equivalent patterns.
Research and memory strategy
- Research Harnesses for Self-Learning AgentsRecord memory-mediated self-learning, evidence ledgers, reproducibility, and static completeness sweeps without turning UAIX into a runtime.
- Agentic Harness Strategies And UAIKeep live tools, retries, approvals, runtime memory, and traces in the harness while UAIX preserves reviewed evidence.
- OpenAI Follow Goals with UAIX Memory FilesUse Codex Goal mode or OpenAI structured prompts as configuration-specific runtime state while UAIX files remain durable memory.
- Long-Running Goal ExecutionCompose runtime goals, .uai checkpoints, UAI-1 packets, blockers, handoffs, final reports, and reviewed write-back.
- Hierarchical Goal MemoryFold completed subgoals into source-linked records and retrieve bounded evidence for the next slice.
- Project Handoff Context BudgetKeep hot memory current and move bulky research to cold memory with promotion evidence.
- Memory FirewallQuarantine imported memory before promotion into active project truth.
Capability-adaptive web interaction
- Capability-Adaptive Web InteractionL0-L6 capability ladder, profile IDs, required fields, fallback rules, and validator expectations.
- Limited Browser SupportStatic and structured fallbacks for clients that cannot run JavaScript.
- Advanced Agent SupportTool, workflow, multi-agent, memory, and audited evidence boundaries.
- Browser and Agent ParityRendered UI evidence tied to static and JSON records.
- UAIX vs MCP vs WebhooksChoose record, tool-session, or event-delivery ownership.
- Agent Consent BoundariesApproval rules for auth, posting, tools, memory, coordination, and audit claims.