Skip to content

v0.197.0

Choose a tag to compare

@github-actions github-actions released this 06 Oct 19:31
· 249 commits to main since this release
4e1f105

v0.197.0 — agents see their own updates mid-turn; file writes can't be read half-done

Added

  • The skill index flags skills that need tools the agent doesn't have (#4061). A skill whose tools: frontmatter names a tool this agent lacks is listed with missing_tools="…", so the agent sees the gap before loading it rather than partway through its procedure. The skill stays listed.

Fixed

  • A task the agent just updated no longer shows its old status on the next step (#4060). The agent's working state (tasks, goals, schedules) used to be read once per turn. After a tool runs, it's now re-read for each model call, so an update the agent just made shows up right away. Prompt caching is unaffected.
  • write_file and edit_file can no longer leave a reader looking at an empty file (#4059).
    • Writes go to a temp file beside the target and are swapped in atomically, keeping the file's permissions.
    • A failed write leaves the original untouched.
    • write_file now takes the same per-file lock as edit_file, so the two can't silently undo each other.

Security

  • Slide previews measure every image in a deck, not just those in ppt/media/ (#4057). A large image stored elsewhere in a .pptx used to skip the pixel caps. Every part is now checked by its contents, and an oversized image refuses the slide preview with the reason shown.