Mechanism
When It Triggers · How It Works · What It Produces
Breaks down how the skill works and what it produces, so you can quickly judge whether it fits your scenario and value.
Trigger Timing
When a user wants an end-to-end, minimal-trustworthy reproduction of a README-documented deep learning repository across intake, setup, execution, and reporting.
Working Mechanism
- Reads README and repo signals first, then calls repo-intake-and-plan to pick the smallest documented target.
- Coordinates env-and-assets-bootstrap, analyze-project, minimal-run-and-audit, or run-train as needed.
- Enforces conservative patch rules, pauses for human review on semantic changes, and writes standardized
repro_outputs/bundle.
Outcome
A reproducible, auditable repro_outputs/ package with summary, commands, logs, patch notes, and comparability report showing faithful reproduction and recorded deviations.
Use Cases
Two Typical Use Cases
Verify documented inference of a GAN repo
A researcher clones a novel GAN repo and wants to verify the documented inference demo. Using this skill, the agent reads the README, selects the smallest inference target via `repo-intake-and-plan`, bootstraps env with `env-and-assets-bootstrap`, runs via `minimal-run-and-audit`, and writes `repro_outputs/`. The researcher gets auditable evidence and knows exactly what assumptions or deviations occurred without silent changes.
Reproduce paper eval before extension work
A engineer must reproduce a paper's evaluation protocol from its official repo before building extensions. The skill orchestrates intake, setup, trusted eval execution, and optional `paper-context-resolver` for conflicts, then records comparability and patches. The engineer receives a standardized bundle proving faithful reproduction and a clear list of human decision points, reducing onboarding risk.
Skill File