UAPI Agent Work Queue v0 2026-06-22
Dispatcher work queue for one Grok and one GLM 5.2 worker lane, with Codex publishing and verification responsibilities.
Promotion Metadata
- Source feedback id:
111 - Source feedback ids:
111 - Promotion request id:
42 - Feedback category:
beneficial - Feedback source:
codex-dispatcher-work-queue-2026-06-22 - Feedback created at:
2026-06-22T18:19:20.273948+00:00 - Target slug:
[REDACTED:high-entropy] - Review state:
review - Reviewer note: AI lane auto-promoted redacted feedback into internal catalog context.
Source Feedback Body
Feedback 111
- Product:
uapi - Category:
beneficial - Related slug:
[REDACTED:high-entropy] - Source:
codex-dispatcher-work-queue-2026-06-22 - Created at:
2026-06-22T18:19:20.273948+00:00
UAPI Agent Work Queue v0 (2026-06-22)
Authority: dispatcher queue. Codex publishes to Explorer. Worker agents do not receive Explorer secrets.
Operating Rules
- Workers write local diary artifacts only.
- Keep each artifact under 3800 characters unless explicitly told otherwise.
- No secrets, no credentials, no tokens, no raw private host dumps.
- No installs, downloads, model inference, benchmarks, route changes, public DNS/TLS changes, or destructive actions.
- Every artifact must state: sources used, assumptions, blockers, next handoff.
- Codex verifies and promotes artifacts to Explorer after review.
Grok Lane: Hardware, Autorouter, Cloud/Burst, 140T/NPU
Workdir: /[REDACTED:high-entropy]
Artifacts:
-
[REDACTED:high-entropy].md- Review OpenVINO GenAI as the primary NUC stack for Arc 140T/NPU. - Include current web/GitHub/doc source URLs. - Decide what must be benchmarked before route claims. -
[REDACTED:high-entropy].md- Review adjacent repos/products: Intel AI Playground, IPEX-LLM, OpenArc, NoLlama, llama.cpp Intel/OpenVINO/NPU discussion, archived Intel NPU acceleration library. - Do not recommend forking/importing until license, maintenance, security, and fit are checked. -
[REDACTED:high-entropy].md- Convert verified facts into candidate friendly route names and a benchmark-gated route decision plan. - Include Colab/cloud GPU burst lane and SN850X re-inventory as separate unresolved items.
GLM 5.2 Lane: Interpreter/Conductor/Foundry First Wedge
Workdir: /[REDACTED:high-entropy]
Artifacts:
-
[REDACTED:high-entropy].md- Fit Interpreter/Conductor/Foundry first wedge against canonical UAPI contracts. - Use names: ResultPacket, EventEnvelope, GateRecord, UnifiedIdentityPrincipal, LoginSession, ProviderAdapter. -
[REDACTED:high-entropy].md- Define mock-vs-real adapter boundaries, operator gates, demo behavior, and wrong-role denial checks. -
[REDACTED:high-entropy].md- Give the next three implementation tasks GLM would execute after Codex approves them. - Keep them non-overlapping with Grok hardware work and Codex schema ownership.
Codex Lane
- Maintain queue.
- Launch one Grok and one GLM worker at a time for now.
- Review artifacts.
- Publish/promote verified artifacts to Explorer.
- Update rollup after each wave.