Version localizada en revision
Esta pagina aun no esta disponible como norma localizada revisada. Use la fuente inglesa autoritativa para lenguaje normativo, interpretacion de esquemas, obligaciones de implementacion y limites de soporte.
Cómo usar esta página
Use this guide to keep long-running agent goals interruptible, checkpointed, review-gated, and written back into authoritative UAIX memory rather than hidden runtime state.
Interruptible Goal Work
Checkpoint long-running work without turning UAIX into a runtime
Use this guide to compose .uai memory, Follow Goals overlays, UAI-1 agent packets, checkpoints, blockers, handoffs, final reports, and reviewed write-back for work that may pause or resume.
Runtime
External runtimes execute
Codex, OpenAI agents, local agents, workflow engines, or humans own execution, tools, approvals, traces, retries, and managed runtime state.
Memory
UAIX files stay authoritative
Runtime state becomes durable project truth only after review and write-back into .uai records, docs, code, tests, release notes, roadmap state, or machine artifacts.
Packets
Checkpoint with existing UAI-1 profiles
Use task-status, blocker, handoff, memory-proposal, final-report, and correction packets instead of inventing a new long-running schema.
Boundary
No hosted orchestration claim
This guide is not a scheduler, hosted memory database, official adapter, SDK, CLI, certification program, automatic repository writer, hosted importer, or automatic sync surface.
Use beside
Follow Goals GuideRuntime objective overlay for environments that use /goal or structured goal prompts.Traspaso de proyectoAuthoritative repo-local memory and handoff boundary..uai File GuideTyped memory files used for checkpoints and write-back.OpenAI/Codex GuideUse OpenAI and Codex runtimes without moving durable memory into the runtime.Coding Agents GuideCode-bearing resume rules and next-recursive prompt boundaries.Context Budget GuideHot/cold memory promotion and compaction rules.Asistente de paquete de memoria de IAConfiguration-specific package guidance for Follow Goals, multi-agent communication, and code-bearing continuation.ValidadorUse current validator behavior only where checks are actually implemented.Registro de cambiosDated public release trail for the guide-level update.
/goal Finish the next bounded work chunk using authoritative UAIX memory, not the last chat turn.
Success criteria:
- Complete the bounded subtask or emit a blocker.
- Run targeted checks or name skipped checks with reasons.
- Record checkpoints, blockers, handoffs, memory proposals, and final evidence with existing UAI-1 profiles where applicable.
- Write accepted facts back to authoritative .uai files after review.
Feature/improvement focus:
- Keep long-running work interruptible, review-gated, and resumable.
Code/source changes:
- Regenerate next-recursive-prompt only after authoritative write-back for code-bearing scopes.
Automated tests/checks:
- Use .uai/test-plan.uai, focused touched checks, final-report evidence, and skipped-check reasons.Use this when a receiver can run /goal, while durable memory remains in UAIX files.