Skip to content

fix(passthrough): close a capped streaming turn as truncated, not as an error #926

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
materemias wants to merge 7 commits into
base: main
Choose a base branch
from
Open

fix(passthrough): close a capped streaming turn as truncated, not as an error #926

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
4a45692
fix(passthrough): close a capped streaming turn as truncated, not as …
materemias Sep 3, 2026
f4b60b2
fix(passthrough): reject empty capped stream turns
materemias Sep 3, 2026
35a0a50
fix(passthrough): preserve capped stream telemetry
materemias Sep 3, 2026
cd42718
fix(passthrough): harden capped stream accounting
materemias Sep 3, 2026
6b8ce36
fix: correlate fresh sessions in passthrough telemetry
materemias Sep 3, 2026
fc95383
fix: correlate capped passthrough sessions
materemias Sep 3, 2026
514caca
fix(passthrough): reissue a capped turn that produced nothing
materemias Sep 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions docs/configuration.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -481,6 +481,14 @@ The cap is lifted for the cases that genuinely need the SDK to keep going — de
> `/telemetry/logs` is what confirms the session was preserved. And a turn that
> hits the cap having produced content but no forwardable tool call is reported
> as `stop_reason: "max_tokens"` (truncated) instead of failing the request.
>
> A capped turn that produced *nothing* — no wire event, no captured tool call —
> never reached the tool boundary the cap exists to stop at, so the single turn
> bought nothing and cost the whole request. Meridian reissues that turn once
> with the cap lifted (`single_turn_cap_lifted` in `/telemetry/logs`), which is
> safe precisely because nothing had reached the client yet. An explicit
> `MERIDIAN_PASSTHROUGH_MAX_TURNS` suppresses the reissue too: the budget is
> then the operator's, and a second identical attempt would only spend a turn.

### Silent turns

Expand Down
Loading
Loading