explore-code
Try small model changes apart from your baseline.
What it does
Makes isolated exploratory changes to deep learning research code and records their rationale, rollback, and unverified candidate status.
When to use it
Use when the researcher explicitly authorizes isolated exploratory changes such as module transplants, backbone adaptations, or LoRA and adapter insertion.
How to use it
Give it an authorized exploratory change in a project, with the current research identifier and experiment branch label; it makes isolated candidate changes and writes records in explore_outputs/.
What you provide
- current research identifier
- experiment branch label
- An existing project
What you need · 4
The bundled planning and output scripts use Python 3.
Uses this skill's shared operating and research-rigor principles; the output writer also looks there for its bundled shared writer.
May hand off execution to this skill; the handoff is permitted rather than required.
May hand off execution to this skill as an alternative to minimal-run-and-audit; the handoff is not required.
Access · 0
No additional actions listed.