Skip to content
Get packs

Reasoning Trace Optimizer

Lets an agent debug and improve other AI agents by analyzing their reasoning traces and prompts

Source Download skill
Repository
muratcankoylan/Agent-Skills-for-Context-Engineering
Stars
18,282
Licence
MIT
Files
15 files

What it does

This skill teaches an agent to examine how an AI agent reasoned through a task, looking for patterns such as lost goals, tool confusion, or premature conclusions. It can then suggest a better system prompt and, in its full form, run repeated capture-analyze-optimize cycles. It is built around MiniMax M2.1 and its interleaved thinking, though the SKILL.md says the session analysis mode can work with other models.

Great for

  • Developers building or debugging agents that call tools repeatedly
  • Teams who want to turn failed agent runs into prompt fixes and shareable notes
  • Anyone experimenting with MiniMax M2.1 reasoning traces

Skip it if

  • Non-developers who just want a writing or research helper; the main modes need a Python package, API keys and scripts that are not included in the file list (the package reasoning_trace_optimizer and rto CLI are referenced but not shipped here)
  • Production systems where automatic prompt rewrites are applied without human review, since the optimizer can change prompts and generated skills should be reviewed before sharing
  • Anyone who does not want their task data or session thinking sent to a third-party API, because captured runs are sent to the MiniMax endpoint

Try asking

  • “Analyze why my agent kept calling the wrong tool after three turns and suggest prompt fixes”
  • “Analyze this session's thinking blocks and tell me where I drifted from the original instructions”
  • “Run an optimization loop on my research agent prompt and stop when the score reaches 80”
  • “Turn my optimization artifacts into a reusable skill for web search best practices”
Step-by-step guide: how to use Reasoning Trace OptimizerInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 15 original files, the MIT licence and an attribution note.

    interleaved-thinking.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip interleaved-thinking.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip interleaved-thinking.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Keep full response history including thinking blocks when capturing traces, as the skill recommends
  • Review any generated skill or optimized prompt before sharing or deploying it
  • Set a small max_iterations value first, since each optimization iteration uses significant tokens

What's inside

  • docs/agentthinking.md
  • docs/interleavedthinking.md
  • docs/m2-1.md
  • examples/01_basic_capture.py
  • examples/02_tool_usage.py
  • examples/03_full_optimization.py
  • generated_skills/comprehensive-research-agent/references/optimization_summary.json
  • generated_skills/comprehensive-research-agent/references/optimized_prompt.txt
  • generated_skills/comprehensive-research-agent/references/patterns_found.json
  • generated_skills/comprehensive-research-agent/SKILL.md
  • optimization_artifacts/final_prompt.txt
  • optimization_artifacts/iteration_1/analysis.txt
  • optimization_artifacts/iteration_1/optimization.txt
  • README.md
  • SKILL.md

Redistributed unmodified from muratcankoylan/Agent-Skills-for-Context-Engineering under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.