analyze-project
Use this as the Rigor Analyze / Rigor Audit read-only skill. The installed slug remains analyze-project for compatibility.
Use the shared operating principles in ../../references/agent-operating-principles.md; this skill should guide read-only analysis without constraining the model's project-specific reasoning.
When to apply
- The user wants to understand a deep learning repository before changing it.
- The user needs a map of model structure, training entrypoints, inference entrypoints, and config relationships.
- The user wants conservative suggestions about likely insertion points or suspicious implementation patterns.
- The user explicitly wants read-only analysis and not heavy execution.
When not to apply
- When the main task is to execute a failing command or debug a traceback.
- When the user wants environment setup or asset download only.
- When the user wants speculative adaptation or broad exploratory patching.
- When the task is a general literature summary without repository analysis.
Clear boundaries
- This skill is read-mostly.
- It may run lightweight static inspection helpers.
- It does not patch repository code.
- It does not own final reproduction outputs.
- It should mark suspicious patterns as heuristics, not confirmed bugs.
Output expectations
analysis_outputs/SUMMARY.mdanalysis_outputs/RISKS.mdanalysis_outputs/status.json
Notes
Use references/analysis-policy.md and the shared references/research-pitfall-checklist.md.
