Iterate

Run iterative research loops until convergence criteria are met.

The iterate workflow executes multi-loop research instead of one-pass generation.

How it works

  1. Starts with a planning and baseline pass.
  2. Runs loop cycles: Research -> Review -> Improve.
  3. Tracks confidence and unresolved contradiction counts each loop.
  4. Stops when thresholds are met or max loops are reached.

Real examples

/iterate "agentic coding benchmarks reproducibility"
bohr iterate "agentic coding benchmarks reproducibility"

Expected output

  • outputs/<slug>-iterative-brief.md
  • outputs/<slug>-iteration-log.md

Iteration log includes per-loop decisions and convergence reason.

Customization

  • Set confidence threshold (for example, >= 0.8)
  • Set contradiction target (for example, zero critical unresolved contradictions)
  • Set max loop count and verification strictness per loop