Overview
Add a stage that uses a strong model (e.g., Opus) to review the codebase and surface additional signals of reachability, such as likely entry points and external data flows.
Motivation
Existing reachability analysis can miss indirect or non-obvious paths, especially those involving unconventional entry points or external inputs.
Proposed Approach
- Introduce an LLM review stage that scans the codebase for:
- Potential entry points
- External content ingestion (APIs, file I/O, network sources)
- Other indicators of possible reachability
- Aggregate findings into a list of reachability signals to augment existing logic
Expected Benefits
- Improved coverage of indirect and edge-case execution paths
- Reduced likelihood of missed entry points
- Stronger overall recall
- This stage complements, not replaces, current reachability analysis
Overview
Add a stage that uses a strong model (e.g., Opus) to review the codebase and surface additional signals of reachability, such as likely entry points and external data flows.
Motivation
Existing reachability analysis can miss indirect or non-obvious paths, especially those involving unconventional entry points or external inputs.
Proposed Approach
Expected Benefits