--- name: repair-build-conveyors description: Drive the three independent Stephanos source-building programmes from exact live evidence, escalate missing capabilities, and prove physical pickup instead of just dispatch. --- # Repair and commission the Stephanos builders ## Mission Run on explicit operator repair intent for goals #2961 (Original Goal Conveyor), #3103/#3109 (Rescue Island), and #3096 (Seed Canopy Parallel Forge). Respect these as **three independent execution domains** with no shared queues, claims, scheduling ownership or original leases. Use `shared/agents/sovereignGoalRecoveryTriageV1.mjs` as a **read-only gap planner**, never an admission or credential authority. ## Bounded fast transport Use `repair-sovereign-fast-bridge` first. Prefer local Sovereign MCP when available; in iPad/cloud chats route through the already installed owner-authenticated GitHub #2808 command mailbox, with a fresh protected `main` SHA and one correlation ID. Start with `visibility-snapshot` and read terminal receipts. If full canonical source protection is held, do not run `repair-goal-builder-flow` repeatedly against the same historical receipt error. ## Independent work-conserving routes **Original:** Existing #2962/r7 lease `critical-2962-elastic-goal-r7-lease` lacks authentic historical terminal execution custody. Install/commission *outside the source checkout* a qualified Windows OS witness, separate protected operator approval and true OS denial of original out-of-job writers, old tasks, WMI/COM and source ACL access. Follow `commission-original-source-custody`. Preserve history as UNKNOWN, verify physical exclusion, and use the original guarded atomic lease compare-and-release only with real signed evidence. No stop-process guesses, fake receipts or force release. **Rescue Island:** Continue authenticated owner GitHub reader PR #3111, already protected merged. Acquire genuine independent OS source-ownership/claim/scope/gaming proof under a reviewed Island-owned adapter before invoking existing bounded `rescueIslandFirstCommitV1.mjs`. Keep old dirty checkout, never share Goal Dashboard or Seed queues. Execute exactly one unclaimed safe pilot #3109 with real SELECTED / CLAIMED / PHYSICAL_PICKUP / EDITED / TESTED / ISOLATED_COMMIT receipts. Publish via distinct PR, independent review and protected merge. Only then clone worker 02 and qualify unattended refills. **Seed Canopy:** PR #3097 is separate and has unresolved safety findings. Keep `SEED_PARALLEL_ROUTER_GOVERNOR_CUSTODY_IMPLEMENTED = false` until a reviewed adapter genuinely registers, stops and independently verifies *all* llama.cpp router processes and GPU/VRAM residency on gaming transition. Never use HTTP abort as model eviction. Resolve source worktree identity, hooks, mutable receipts, symlink, and Git config hardening with tests and specialist review. No high-VRAM launch that harms Quest/VR or flat gaming. **Native capacity:** PR #3099 proves task installation is not enough. Inspect the exact terminal result of `INSTALL_STEPHANOS_NATIVE_CAPACITY_PUBLISHER`; require the separately signed same-head candidate and fresh readback. An accepted mailbox request is not install success and must not be duplicated while in flight. ## Autonomous blocker progression For each HOLD, emit a specific capability build or vetted skill-import request with `{owner: source programme, exact blocked gate, required evidence, scope, proof test, safety review}`; choose another *independent* admitted task instead of stalling all lanes. Skills cannot create OS principals, privileged signers, permissions, admission rights, approved workloads or protected merge authority. Use Source PR → tests → independent exact-head review → operator-approved protected merge → Battle Bridge main sync → physical acceptance proof. Final green ONLY after authentic source DIFF/TEST/COMMIT, independent protected merge, runtime observation and **two unattended refills**. No implication of asynchronous work from a chat turn.