Skip to content

Report safe token counting diagnostics separately from compaction - #14

Merged
Brian Krabach (bkrabach) merged 1 commit into
mainfrom
codex/count-recovery
Oct 8, 2026
Merged

Brian Krabach (bkrabach) merged 1 commit into
mainfrom
codex/count-recovery

Conversation

@bkrabach

@bkrabach Brian Krabach (bkrabach) commented Oct 8, 2026 •

Copy link
Copy Markdown
Collaborator

Distinguish a provider token-counting failure from a compaction or foreground generation failure. Project only its category, HTTP status, request ID, bounded attempt count, and manual retry guidance into the generation failure event. Never copy SDK bodies or prompt data.

This does not add automatic turn or generation retries. The provider exception remains non-retryable by the outer execution loop; its separate diagnostic describes whether a later explicit continuation may succeed.

Validation in the project-owned DTU: 139 loop tests passed and the final 10 failure-projection tests passed. The provider is optional and is not imported by the core loop.

Related changes: provider #123, context #21, loop #14, Unified #468. Deploy the context and loop changes before or alongside the provider change.

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