When the baseline stage sees a red gate it assumes a pre-existing code defect and dispatches a builder agent to fix it by TDD. When the failure is environmental -- no Docker for the test harness, or a provider returning 402 for the fix run itself -- nothing in the worktree can fix it, yet the run still spends a full agent budget (observed: 34 iterations, ~1.29M tokens, 5 changes, then parked) and then reports the gate as a task-level failure. A provider error during the fix run exits the stage with the provider message instead of the real gate cause. Classify the failure at the producer: an infrastructure/credential cause must park (or report) with that cause and must not dispatch a builder, and the fix run's own provider failure must not mask the gate output.
Acceptance criteria
a gate failing for an environmental reason parks with that reason and dispatches no builder; a real code failure still dispatches one; a provider failure during the fix run is reported as such alongside the gate output
Close reason
Closed
Imported from beads archie-core-rwy6 · GitHub #1466 · type bug · priority P2 · created 2026-10-01 · closed 2026-10-02
When the baseline stage sees a red gate it assumes a pre-existing code defect and dispatches a builder agent to fix it by TDD. When the failure is environmental -- no Docker for the test harness, or a provider returning 402 for the fix run itself -- nothing in the worktree can fix it, yet the run still spends a full agent budget (observed: 34 iterations, ~1.29M tokens, 5 changes, then parked) and then reports the gate as a task-level failure. A provider error during the fix run exits the stage with the provider message instead of the real gate cause. Classify the failure at the producer: an infrastructure/credential cause must park (or report) with that cause and must not dispatch a builder, and the fix run's own provider failure must not mask the gate output.
### Acceptance criteria
a gate failing for an environmental reason parks with that reason and dispatches no builder; a real code failure still dispatches one; a provider failure during the fix run is reported as such alongside the gate output
### Close reason
Closed
---
_Imported from beads `archie-core-rwy6` · GitHub #1466 · type bug · priority P2 · created 2026-10-01 · closed 2026-10-02_
sam
added the mediumbug labels 2026-10-27 13:40:52 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
When the baseline stage sees a red gate it assumes a pre-existing code defect and dispatches a builder agent to fix it by TDD. When the failure is environmental -- no Docker for the test harness, or a provider returning 402 for the fix run itself -- nothing in the worktree can fix it, yet the run still spends a full agent budget (observed: 34 iterations, ~1.29M tokens, 5 changes, then parked) and then reports the gate as a task-level failure. A provider error during the fix run exits the stage with the provider message instead of the real gate cause. Classify the failure at the producer: an infrastructure/credential cause must park (or report) with that cause and must not dispatch a builder, and the fix run's own provider failure must not mask the gate output.
Acceptance criteria
a gate failing for an environmental reason parks with that reason and dispatches no builder; a real code failure still dispatches one; a provider failure during the fix run is reported as such alongside the gate output
Close reason
Closed
Imported from beads
archie-core-rwy6· GitHub #1466 · type bug · priority P2 · created 2026-10-01 · closed 2026-10-02