All pull requests

Pull requests

• fix(usage): preserve cached input tokens across provider bridges

  #5672·aspire488 opened Oct 4, 2026·

  ·
  1

• fix(tools): convert structured validation failures to parsing errors

  #5671·aspire488 opened Oct 4, 2026·

  ·
  1

• fix(google): prefer explicit API keys over ambient ADC

  #5670·aspire488 opened Oct 4, 2026·

  ·
  1

• Let tasks and samples carry a description, recorded in the log and dataframes

  qualified
  #5667·dragonstyle opened Oct 3, 2026Collaborator·

  ·
  1

  1

  

• Wire-type modules load without the rest of inspect_ai

  #5664·epatey opened Oct 2, 2026Collaborator·

  ·
  1

• fix: support eval logs with prefix-scoped S3 access

  qualified
  #5660·rasmusfaber opened Oct 2, 2026Contributor·

  ·
  

• Fix Antigravity host tool calls being denied as unproposed by the sandbox agent bridge

  qualified
  #5658·sjawhar opened Oct 2, 2026Contributor·

  ·
  

• Fix remote MCP tools running without approval and bridged tool results skipping output limits

  qualified
  #5654·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Design: Working time under concurrency within a sample

  #5653·ransomr opened Oct 1, 2026Collaborator·

  ·

• Fix tool approval deciding on a different call from the one that runs

  #5652·ransomr opened Oct 1, 2026Collaborator·

  ·

• Fix approvers' modify decisions running the wrong call or logging the wrong arguments

  qualified
  #5651·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Agent bridge follows the eval's configuration for service tier, storage, truncation and tool options

  qualified
  #5650·ransomr opened Oct 1, 2026Collaborator·

  ·
  1

  

• Fix copy-out cancellation test failing on Python 3.14

  qualified
  #5649·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix model output cache reading, writing and deleting files outside the cache directory

  qualified
  #5648·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Stop Mistral providers downloading image URLs from model output

  qualified
  #5647·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix batch mode sending one request's headers with other requests in the batch

  qualified
  #5645·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix costs and cost limits pricing requests at the requested model instead of the model that served them

  qualified
  #5644·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix usage undercounted when a provider call sends several billed requests

  #5643·ransomr opened Oct 1, 2026Collaborator·

  ·

• Fix sample limits being ignored when exceeded inside tool calls and MCP sampling

  #5642·ransomr opened Oct 1, 2026Collaborator·

  ·

• Fix HF provider generating concurrent requests with another model's settings

  qualified
  #5641·ransomr opened Oct 1, 2026Collaborator·

  ·
  1

  

• Fix model calls being sent and billed after a token or cost limit is reached

  qualified
  #5640·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix Perplexity responses picking up another request's citations and usage

  qualified
  #5638·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Fix Google audio, video and document content sent with another upload's MIME type

  qualified
  #5637·ransomr opened Oct 1, 2026Collaborator·

  ·
  

• Test framework writes across coreutils variants

  qualified
  #5636·rasmusfaber opened Oct 1, 2026Contributor·

  ·
  

• Fix model calls in multiprocess Scout scans

  qualified
  #5635·rasmusfaber opened Oct 1, 2026Contributor·

  ·