Skip to content

CL-8978: prevent recoverable subagent failures from stalling parent - #1171

Open
TheGreatAxios wants to merge 3 commits into
mainfrom
cl-8978-prevent-recoverable-subagent-failures-from-stalling-parent
Open

TheGreatAxios wants to merge 3 commits into
mainfrom
cl-8978-prevent-recoverable-subagent-failures-from-stalling-parent

Conversation

@TheGreatAxios

Copy link
Copy Markdown
Collaborator

Implements CL-8978: a worker that fails after returning an id with a transient provider fault (retryable/timeout/429-overload) now settles wait-terminal failed with a machine-readable continuable marker plus single-successor guidance, instead of leaving the parent idle-with-fleet or stall-aborting.

What changed:

  • src/inference-error-message.ts: isRecoverableProviderFailureCategory allowlist (retryable/timeout); credential/quota/context-overflow win as fatal.
  • src/subagent/agent-fleet.ts: agent-fleet catch stamps markRecoverable alongside sessions.fail, snapshot projects recoverableFailure, wait_agents description documents the marker.
  • src/subagent/fleet-dry-drive.ts: failed+recoverable entries project continuable:true plus capped continue_with guidance to both wait_agents and mailbox mail.
  • src/subagent/run-recoverable-failure.test.ts: red-then-green coverage (marker, fatal control, mail delivery with send-failure re-arm, in-flight liveness, description).

No auto-retry added; credential/quota/context-overflow handling untouched.

Fixes CL-8978

@linear-code

linear-code Bot commented Sep 25, 2026

Copy link
Copy Markdown

CL-8978

Wait output for a recoverable child failure tells the parent to spawn
one successor with the same brief. The standing orchestration rule still
refused any identical brief, so Skywalker would refuse the spawn the
marker asks for.
@TheGreatAxios
TheGreatAxios force-pushed the cl-8978-prevent-recoverable-subagent-failures-from-stalling-parent branch from 7182629 to 0666b74 Compare September 26, 2026 21:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant