Tool synthesis systems let agents create new tools from external resources (codebases, papers, APIs), typically in a closed loop:
- Generate tool code.
- Run tests or validation.
- Repair failures.
- Register tool for future calls.
Why It Matters
This shifts tool coverage from a manually curated catalog to a partially self-extending system—useful, but risky without strong sandboxing.