Skip to content

Releases

v0.3.0

Choose a tag to compare

@key4ng key4ng released this 31 Mar 19:08

What's New

Subagent support

  • New spawn_subagent tool for delegating subtasks to different models
  • Subagents run in isolation with configurable tool access (defaults to read-only)
  • Parallel execution via run_subagents_parallel()
  • Recursive spawning prevented by design
  • Includes ensemble-review example plugin

Bug fixes

  • Fix OCI Responses API tool call IDs — generate proper fc_-prefixed IDs instead of appending _fc to call_* IDs
  • Fix shallow copy of skill tool definition — use deepcopy to prevent description accumulation across calls
  • Fix token budget early-exit — now checks budget after tool execution instead of before, so pending tool calls aren't skipped
  • Pass timeout to subagent execution — subagents now respect the parent's timeout setting

Full Changelog: v0.2.1...v0.3.0

v0.2.2

Choose a tag to compare

@key4ng key4ng released this 14 Mar 01:11

What's New

  • Native OCI Responses API for xai.* and openai.* models — direct HTTP + SecurityTokenSigner, bypassing litellm
  • All other models (google, meta, non-OCI) fall back to litellm chat completions
  • Removed _patch_oci_config hack

What's Changed

  • feat: native OCI Responses API for xai/openai models by @key4ng in #1

New Contributors

  • @key4ng made their first contribution in #1

Full Changelog: v0.2.1...v0.2.2

v0.2.1

Choose a tag to compare

@github-actions github-actions released this 08 Mar 07:48

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Choose a tag to compare

@github-actions github-actions released this 08 Mar 05:17

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 08 Mar 03:37