The planner that decides whether a turn continues an existing server-side message thread has a new input, creditRetryStateless. When it is set, the planner does not continue the thread. The request is sent stateless instead, meaning it carries the conversation itself rather than pointing at a thread stored on the server.
This joins the other situations that already force a stateless request: holds from the model, relay, classifier and drop paths. The new value is also recorded on the tengu_tether_live_outcome telemetry event.
Why
This adds one more case where Claude Code falls back to a plain stateless request. Nothing changes in what you type or see. The difference is in how the turn is sent to the server.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtThe finding does not say what sets `creditRetryStateless`.
Anthropic's release notes agreeImproved screen-reader output in /mcp: a disabled server is read as "off" instead of "pending"
