think — record reasoning without acting

think is a deliberate no-op. It writes the supplied thought into the run’s transcript and returns immediately. No environment access, no side-effects — just a structured place to externalise reasoning so the next turn (or a reviewer) can see it.

Schema

Arg	Type	Required	Description
thought	string	yes	The reasoning to record.

Example invocation

{"thought": "The bug is in retry logic. Options: (a) cap retries, (b) backoff jitter. Picking (b) — simpler."}

from chimera.tools.think import ThinkTool

tool = ThinkTool()
result = tool.execute(
    {"thought": "Need to check whether the env var is set before reading."},
    env=None,
)
print(result.metadata["thought"])

Output sample

Thought recorded.

The actual text lives in result.metadata["thought"] and shows up in the StepCost event for the turn.

When to use it

• Plan-and-execute loops — write the plan, then execute.
• Reflection / critique — capture self-correction without polluting the user-facing output.
• Long multi-step traces — anchor decisions for later review.

See also

• Loops — PlanAndExecute, Reflexion, TreeOfThought.
• todo — track multi-step work explicitly.