investigate

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes local scripts to perform advanced codebase analysis.
  • It executes python .claude/scripts/code_graph with several subcommands (such as connections, trace, and query) to map code dependencies and caller graphs.
  • It executes node .claude/scripts/ai-signal-scan.cjs to identify AI-related patterns or artifacts in the source code.
  • These scripts are located in the internal .claude/scripts/ directory and are used for their intended analytical purposes.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data which could contain malicious instructions.
  • Ingestion points: Processes user-supplied bug descriptions in --mode=debug and reads arbitrary source code files from the repository during the discovery and analysis phases.
  • Boundary markers: Implements strict "Evidence-Based Reasoning" requiring file:line citations for every claim. It also uses an "End-to-Start Debugger Trace" protocol to verify findings against actual code behavior.
  • Capability inventory: The skill has the capability to read filesystem contents and execute specific local Python and Node.js scripts. It is explicitly restricted to a "READ-ONLY" posture for source code.
  • Sanitization: Findings are synthesized through structured phases (Flow Mapping, Analysis, Synthesis) which serves as a conceptual processing boundary, although no explicit input sanitization code is present.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 10:11 AM