UAI-1 is the current public message standard for structured AI-to-AI communication.
UAIX shield mark Public interoperability charter UAIX.org Universal Artificial Intelligence Exchange Public standards publication
Current UAI-1
  • Current version UAI-1
  • Plain-English definition Open message format for auditable AI-to-AI exchange
  • Current attribution Michael Joseph Kappel, MCP
  • Start here Get Started, Project Handoff, UAI-1, validator, roadmap, and changelog

Guides

Using UAIX Memory Files with OpenAI Follow Goals

Default-off configuration-specific runtime overlay guidance for using Codex Goal mode or OpenAI structured prompts while UAIX memory files remain the reviewed durable source of truth.

  • Record UAIX-DOC-3258
  • Path /en-us/guides/project-handoff-openai-follow-goals/
  • Use Canonical public record

Document status

Public standards page Published on UAIX as part of the current public standards record
Code
UAIX-DOC-3258
Surface
Guides
Access
Public and linkable

How to use this page

Use this guide when Codex Goal mode or an OpenAI structured prompt should consume UAIX memory as configuration-specific runtime state while durable writeback remains review-gated.

Use beside

OpenAI GuideAI Memory Package WizardProject HandoffContext Budget Guide

Default-Off Runtime Goal Overlay

Use Follow Goals without moving durable memory into runtime state

UAIX does not require Follow Goals as a universal runtime. When a receiver environment uses Codex /goal or an OpenAI structured prompt block, generate runtime guidance from UAIX memory files; for /goal receivers, .uai/exports/follow-goal.json is configuration-specific required to unify the system.

Runtime

Goal state stays temporary

Codex /goal state and OpenAI prompt state can focus a run only in environments that use them, but they do not become accepted project memory by themselves.

Completion

Show goal prompts only for goal-origin work

When an actual /goal or explicit goal-prompt run completes, the final response should include a prominent, sectioned Suggested Next /goal Prompt whose copy-paste block starts with /goal and uses goal wording/functionality. Ordinary non-goal work still gets a regular Suggested Next Prompt, even when Follow Goals export support is present.

Memory

UAIX files stay authoritative

Startup packet, receiver brief, constraints, test plan, progress, and short-term memory remain the reviewed durable source of truth.

Boundary

Configuration-specific support

The overlay is default-off generated guidance and JSON export for chosen receiver configurations. .uai/exports/follow-goal.json is required when the receiver environment uses /goal, but this is not an official OpenAI adapter, SDK, CLI, hosted importer, certification, or native API goal field.

Use beside

OpenAI GuideMain OpenAI/Codex Project Handoff framing.AI Memory Package WizardEnable Alternative: Follow Goals in Step 7 only when the receiver runtime uses it.Project HandoffCore repo-local durable memory pattern.Context Budget GuideKeep handoff startup context compact across runtime sessions.RoadmapCurrent versus planned adapter, SDK, CLI, and certification boundaries.
BoundaryRuntime versus durable memory
UAIX does not require Follow Goals as a universal runtime. Required when the receiver environment uses /goal: .uai/exports/follow-goal.json. Codex /goal and OpenAI structured prompts are configuration-specific runtime overlays; reviewed UAIX files remain durable project memory. When an actual goal completes, show a prominent Suggested Next /goal Prompt whose copy-paste block starts with /goal and includes Goal, success criteria, constraints, verification, Feature/improvement focus, Code/source changes, and Automated tests/checks. Ordinary non-goal work still ends with regular Suggested Next Prompt.

Use this line when an OpenAI goal prompt risks sounding like accepted memory.

Follow Goals can help an OpenAI runtime stay focused, but UAIX memory files remain the durable source of truth. Use this guide when a receiver wants Codex Goal mode or an OpenAI structured prompt to carry a measurable objective that was derived from a reviewed UAIX Project Handoff bundle.

This page is intentionally narrow. It describes a default-off, configuration-specific runtime overlay for OpenAI-centered work. It does not turn UAIX into an OpenAI runtime, official adapter, SDK, CLI, hosted importer, repository writer, certification program, or native API goal-field provider.

What This Page Is For

  • Use it when a handoff should include one explicit runtime goal, success criteria, constraints, verification surfaces, and reference files.
  • Use it after the receiver has the normal Project Handoff files and before runtime-specific work begins.
  • Use it to keep the OpenAI runtime objective separate from reviewed durable memory until results are accepted and written back.
  • Use the broader OpenAI and Codex guide when you need the complete Project Handoff pattern for OpenAI agents.

Layer Split

Layer Job Durability
UAIX memory files Preserve current truth, constraints, decisions, verification plans, receiver briefs, and reviewed writeback. Durable project memory under review.
Codex Goal mode Keep a Codex app, IDE, or CLI session pointed at one persistent objective, with planning first when selected. Runtime session state, not accepted memory by itself.
OpenAI structured prompt Give an OpenAI API or chat workflow a clear Goal / Success criteria / Constraints / References / Verification block. Runtime prompt state, not accepted memory by itself.
.uai/exports/follow-goal.json Carry the runtime overlay as structured export material when the wizard setting is selected. Configuration-specific required for receiver environments that use /goal; generated from the handoff, not a new durable memory source.

Wizard Setup

  1. Open the AI Memory Package Wizard.
  2. Configure the normal Project Handoff package first, including startup packet, receiver brief, constraints, test plan, progress, and next recursive prompt where the package family requires it.
  3. In Step 7, open Alternative: Follow Goals. It is off by default.
  4. Choose Codex Goal mode, OpenAI structured prompt block, or both.
  5. Enter a primary goal that fits into Codex goal text and move long detail into the referenced UAIX files.
  6. List success criteria, constraints, verification surfaces, and references explicitly enough that the human can revise the next action before the receiver starts.
  7. When the receiver uses /goal, keep .uai/exports/follow-goal.json in the package as the configuration-specific required unifying record.

Default References

A Follow Goals overlay should point the runtime back to the same durable Project Handoff files the receiver already needs:

  • .uai/startup-packet.uai for startup read order and setup boundary.
  • .uai/receiver-brief.uai for first task, success evidence, and handoff framing.
  • .uai/constraints.uai for hard boundaries and support-claim limits.
  • .uai/test-plan.uai for checks and acceptance evidence.
  • .uai/progress.uai and .uai/short-term-memory.uai for reviewed status and writeback.

Codex Goal Path

For Codex app, IDE, or CLI work, the runtime overlay can produce a short /goal command and a plan-first reminder. Use /plan first when the objective needs refinement, and keep detailed instructions in files when the goal text would become too long.

Code example
/plan Refine the UAIX handoff into a concrete, testable goal before execution.
/goal Complete the receiver task using the generated UAIX memory files, preserve constraints, run targeted checks, and record evidence before stopping.

The Codex runtime goal does not replace AGENTS.md, .uai/readme.human, .uai/startup-packet.uai, or the typed .uai records. It should be derived from them and reported back to them only after review.

OpenAI API Path

For OpenAI API use, treat Follow Goals as a structured prompt block, not a native top-level API goal parameter. Keep the prompt explicit and attach or reference the same files that carry durable memory.

Code example
Role: You are an agent using UAIX files as reviewed durable context.

# Goal
Complete the receiver task with verifiable evidence.

# Success criteria
- Current truth and constraints are summarized before broad work.
- Changed files, checks, skipped checks, blockers, and owners are recorded.

# Constraints
- UAIX memory files remain the durable source of truth.
- Runtime output is not accepted memory until reviewed.

# Verification
- Use .uai/test-plan.uai and .uai/progress.uai as the evidence route.

When the runtime supports file inputs or persistent conversation state, use those OpenAI mechanisms to provide the UAIX references. Do not claim that UAIX publishes a native OpenAI goal field, official importer, or automatic repository writeback.

Review And Writeback

  1. Before work starts, the receiver should summarize current truth, constraints, touchpoints, and targeted checks.
  2. During work, runtime status can be active or paused, but pause reasons and resume conditions should be explicit.
  3. After work, promote only reviewed accepted facts into .uai/progress.uai, .uai/decisions.uai, .uai/short-term-memory.uai, .uai/test-plan.uai, release notes, code, tests, or other authoritative records.
  4. Keep raw runtime traces, prompts, and generated summaries out of durable memory unless they are reviewed, redacted, and summarized for a named purpose.

Conditional Goal-Shaped Completion Handoff

Use goal wording and functionality only when work started from /goal, the human explicitly asked for /goal or a goal prompt, or the last accepted handoff was goal-shaped. Selecting Alternative: Follow Goals as an available runtime export does not by itself convert ordinary continuation into /goal. For true goal-origin work, end with Suggested Next /goal Prompt, not only a generic Suggested Next Prompt, make the copy-paste block start with /goal, and include Goal, Success criteria, Constraints or hard boundaries, Verification, Feature/improvement focus, Code/source changes, and Automated tests/checks so the human can steer the next objective before another agent continues.

Validation Expectations

  • Warn or block when Follow Goals is selected with no primary goal, no success criteria, or no verification surface.
  • Warn when a paused goal lacks a pause reason or resume condition.
  • Warn when the overlay claims official OpenAI adapter support, hosted import, SDK, CLI, certification, runtime orchestration, automatic repository writing, or native Responses API goal support.
  • Keep .uai/exports/follow-goal.json configuration-specific required for /goal receiver environments; universal launch-baseline packages that do not select Follow Goals leave that file out of scope.

OpenAI Docs To Pair With This Guide

Support Boundary

  • Current support: wizard guidance, receiver text, Codex /goal handoff text, structured OpenAI prompt text, and .uai/exports/follow-goal.json as a configuration-specific required export for /goal receiver environments.
  • Not current support: official OpenAI adapter, hosted importer, SDK, CLI, certification, endorsement, runtime orchestrator, repository writer, automatic trace importer, automatic memory promotion, or native Responses API goal field.
  • Durable source of truth: reviewed UAIX memory files, public UAIX pages, release notes, code, tests, and accepted project records.

Long-running work

For extended or interruptible goals, use Long-Running Goal Execution. Follow Goals carries the runtime objective; authoritative .uai files carry durable checkpoints; UAI-1 task-status, blocker, handoff, memory-proposal, and final-report packets carry portable evidence. Runtime output is not accepted memory until reviewed and written back.

Agent Communication Operating Model

Agent runtimes execute. UAIX records the reviewed communication, memory, trust, evidence, and handoff boundary.

Use Agent Communication Operating Model when a human operator, coding agent, AI agent, or runtime implementer needs portable identity, intent, context, acknowledgement, status, blocker, memory proposal, validation evidence, correction, final report, or handoff records. The matching machine-readable assets are published through Schemas, Registry, Examples, and the Validator.

  • UAIX does not provide automatic sync, hosted messaging, runtime orchestration, official adapters, repository write execution, certification authority, or hosted import mechanisms.
  • 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 and other runtimes may consume the published UAIX records as external consumers; UAIX.org does not implement their project-specific runtime features.
  • Durable memory proposals stay separate from task execution records. Do not promote temporary endpoint statuses, session tokens, write tokens, beta platform instructions, local error traces, private keys, or API credentials.