paper-reading-experiment-design
Installation
SKILL.md
Experiment Design
Use this skill to turn a stable method's claims into an experiment plan. Treat the skill as a Claim -> Evidence -> Protocol -> Reviewer Defense converter: it does not complete the method, and it does not begin from benchmark, baseline, or metric lists. Baseline selection and metric selection are part of this workflow, not separate downstream skills.
If the Method Thesis, Mechanistic Claim, target failure, or intervention point is still unclear, route the user back to method commitment instead of inventing experiments for an unstable method.
Output Root
Set {workspace-root} before creating, scanning, or updating artifacts:
- Default
{workspace-root}toworkspace(the repo-localworkspace/directory). - If the user specifies a workspace root, use that path exactly and do not add another
workspace/layer. - If the user provides an existing artifact or workflow path, infer
{workspace-root}from that path and keep related artifacts under the same root. - Do not create generated workflow directories directly at the repository root.
Create or resume one experiment folder at: