All runbooks
runbookuapiredacted view

UAPI Master Program Control Runbook

Master coordination runbook for the UAPI multi-agent build wave.

Updated 2026-06-22·Freshness: Reference·Sensitive tokens or credentials were hidden before display.

UAPI Master Program Control Runbook

Master coordination runbook for the UAPI multi-agent build wave.

Promotion Metadata

  • Source feedback id: 78
  • Source feedback ids: 78
  • Promotion request id: 13
  • Feedback category: consider
  • Feedback source: ai-lane
  • Feedback created at: 2026-06-22T16:31:03.962320+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 78

  • Product: uapi
  • Category: consider
  • Related slug: [REDACTED:high-entropy]
  • Source: ai-lane
  • Created at: 2026-06-22T16:31:03.962320+00:00

title: UAPI Master Program Control Runbook project: uapi kind: runbook state: review tags: [docs, runbook, uapi, program-control, multi-agent] source_plan_title: UAPI Integrated Operating Plan v2

UAPI Master Program Control Runbook

Purpose: coordinate the UAPI mega-project through Explorer as source of truth. This runbook controls handoffs, diaries, gates, audit reporting, and final synthesis across Codex, Grok, and GLM 5.2.

Access contract: Endpoint: http://10.0.0.121:8080. Token: set UAI_EXPLORER_AI_API_KEY out of band. Header uses Bearer token from that variable. Never print secrets. Do not treat MacBook, NUC, Zenbook, Proxmox, Cortex, Beacon, Mesh, Cadence, cloud GPU hosts, or any lab device as local.

Boot sequence for every agent: 1. GET /ready and confirm status=ready, public_safe_mode=true, redaction_boundary=server-side. 2. GET /api/ai/bootstrap and fetch every required accessible primary_context. 3. Fetch this runbook, the main integrated UAPI plan by catalog title, and the assigned agent runbook. 4. Report catalog titles/slugs used, audit_event_ids, readiness checks, changes planned/made, blockers, and feedback submitted.

Work split: - Codex: schema registry, unified login, contracts, integration gates, final synthesis. - Grok: hardware review, autorouter, SN850X, VPS/Cortex host reality, NUC CPU, 140T, NPU, Colab/cloud GPU, friendly route names. - GLM 5.2: Interpreter first wedge, Conductor/Future Foundry gateway behavior, mock-vs-real adapter review, demo acceptance.

Diary rules: each active session maintains an Explorer-visible diary packet or feedback item with agent, runbook title, current task, facts verified, assumptions, audit ids, next action, and blocker. No diary may contain raw tokens, secrets, private keys, or unredacted credentials.

Acceptance gates: - Gate A: source contexts fetched and audit ids recorded. - Gate B: facts separated from assumptions. - Gate C: schema or implementation change has tests/sanity checks. - Gate D: cross-agent contradictions resolved before merge. - Gate E: final runbook/context fetch returns 200 with raw_available_to_client=false.

Continuation rule: when a session nears exhaustion, publish a continuation packet in Explorer with exact next action and audit ids. Do not invent hidden autonomous work. Hourly automation may detect stale work and create a continuation prompt, but must not perform unapproved infrastructure mutation.

Master prompt starter: Use UAI Explorer AI lane only. Fetch /ready, /api/ai/bootstrap, all required primary contexts, this runbook, and your assigned runbook. Work inside your boundary, keep a diary, and report only verified Explorer or explicitly cited web/GitHub facts. Never print secrets.