planning-codebase-restructuring
Pass
Audited by Gen Agent Trust Hub on Jul 9, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates in a read-only mode for the target codebase. It explicitly forbids dangerous operations such as running tests, builds, package installations, formatters, code generators, or any commands that write to the target repository tree.
- [SAFE]: Robust defensive boundaries against Indirect Prompt Injection are implemented. The orchestrator and subagents (such as the reference-assessor) are strictly instructed to treat repository content and fetched web content as data, never instructions. Any instruction-like content encountered in data must be quoted in 'Security notes' and ignored.
- [SAFE]: Network activity is limited and transparent. The skill performs shallow cloning of a user-provided repository URL to a disclosed directory outside the target tree and fetches architecture methodology documentation from a predefined list of well-known reputable sources (e.g., martinfowler.com, owasp.org).
- [SAFE]: A strict 'Summary Contract' and a dedicated 'Plan Reviewer' subagent enforce quality and security. These mechanisms ensure findings are traceable to repository evidence, summaries do not contain unvalidated dumps or instructions, and all proposal actions are approval-gated with safer alternatives provided.
Audit Metadata