Skip to content

feat: help the model reason over recent tool failures, especially repeated failures, instead of only seeing isolated tool outputs.#32

Open
1330482928 wants to merge 2 commits intoMagicCube:mainfrom
1330482928:feat_agent_runtime_control
Open

feat: help the model reason over recent tool failures, especially repeated failures, instead of only seeing isolated tool outputs.#32
1330482928 wants to merge 2 commits intoMagicCube:mainfrom
1330482928:feat_agent_runtime_control

Conversation

@1330482928
Copy link
Copy Markdown
Contributor

This PR introduces the first stage of Agent Runtime Control and Observation.
It adds:

  • normalized tool result envelopes
  • recent tool trace collection
  • recovery hint policies for common failures
  • transcript compaction / summary-first shaping
  • prompt-level tool observation injection before the next model step

The main goal is to help the model reason over recent tool failures, especially repeated failures, instead of only seeing isolated tool outputs.

 - normalized tool result envelopes
 - recent tool trace collection
 - recovery hint policies for common failures
 - transcript compaction / summary-first shaping
 - prompt-level tool observation injection before the next model step
…eated failures, instead of only seeing isolated tool outputs.
@1330482928 1330482928 changed the title feat: feat: help the model reason over recent tool failures, especially repeated failures, instead of only seeing isolated tool outputs. Apr 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant