Skip to content

cloud sessions: no rate-limit telemetry available in-session — agents cannot dynamically scale concurrency #2747

Description

@kyle-sexton

Problem

During the 2026-08-15 session that produced PR #2700, the org's API budget hit sustained 429s: the repo's automated Claude review lanes failed repeatedly with api_error_status: 429 (their own infra comments classified it rate-limit), and the interactive session had no way to observe rate-limit state — the operator's rate-limit guard/statusline exists only in their local global config, not in cloud containers. The operator had to notice degradation manually and instruct the session to pause; the session could only run a fixed conservative concurrency cap (2 subagents) chosen blind.

Ask

Expose rate-limit telemetry to cloud sessions so agents can scale dynamically instead of guessing: options include (a) porting the rate-limit-guard hook/statusline into the cloud environment's setup (fleet cloud-environment component), (b) a lightweight polling surface (e.g. a script or MCP tool reading the org's usage/limits), or (c) at minimum a documented convention for agents to back off on observed 429s from sibling automation (the review-lane infra comments are already machine-readable).

Requested by the operator in-session ("Maybe I'll have to file an issue... so you have access to see what the rate limits are in real time, because then you can kind of dynamically scale").

Provenance

Session 3831ac01-8922-5eec-a8ca-893fe769e590; observed 429 evidence on PR #2700 (runs 31872731086, 31872731063, 31873175085, 31873175066).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions