Open-sourced: Ignat — multi-agent scientific review

We open-sourced Ignat — a collection of Claude Code skills and specialist agents for scientific paper review, cross-disciplinary insight, and systematic invention.

The problem it addresses: single-pass AI review inherits well-documented biases — sycophancy (inflating weak papers) and anchoring (fixating on the first impression). Ignat counters both by design.

v0.1 ships three commands:

  • /review — a 4-round manuscript review: an advocate finds the strengths, then an adversarial critic (positive language banned) attacks every claim, then an independent specialist panel checks in parallel, then a synthesis round cross-confirms. Citation verification and a novelty audit are built in.
  • /insight — six scientists from maximally distant fields (physicist, biologist, engineer, mathematician, earth scientist, computer scientist) read the same paper. A mathematician questions exactly the assumption a chemist takes for granted.
  • /triz — a 3-round TRIZ brainstorm (improve → break → wild) that produces 50–70 ranked hypotheses.

Behind the commands is a panel of seven specialist agents, each with its own lens.

Ignat was born from this project. A project that sets out to design life-like matter needs a physicist, a chemist, a biologist, and an engineer to review the same paper — so we built the tool for ourselves, then realized it would be useful to anyone. It is named after the AI assistant who co-designed it.

MIT licensed. Contributions welcome.

Ignat on GitHub →