academic-repro

Academic Reproduction with AutoForge

Use this skill when the user wants to:

• infer likely papers from a research goal
• reproduce a paper or build a reproduction brief
• extract claims, datasets, metrics, and environment assumptions
• queue a long-running research workflow and monitor it over time

Preferred workflow:

1. If needed, start the durable runtime:
   • autoforgeai daemon start
2. For paper discovery or reproduction artifacts:
   • autoforgeai paper infer "<goal>"
   • autoforgeai paper reproduce "<goal>" --top-k 5 --pick 1
3. For long-running work, prefer queue/watch over one-shot execution:
   • autoforgeai queue "<research objective>"
   • autoforgeai watch <project_id>
   • autoforgeai msg <project_id> "<new constraint or direction>"
4. Preserve and summarize the resulting artifacts:
   • candidate.json
   • reproduction_brief.md
   • generation_prompt.txt
   • paper_signals.json
   • verification_plan.json
   • environment_spec.json
   • run_manifest.json
   • repro_report.json

Operational guidance:

• Use the daemon path when the research workflow may take more than one interactive Claude turn.
• Prefer evidence and artifact paths over vague summaries.
• Call out whether the result is reproduced, partially reproduced, or not reproduced.
• If the user wants interactive correction mid-run, use msg or unpause rather than restarting from scratch.