rightcode-codeprobe

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill implements a robust 'READ-ONLY' constraint, explicitly forbidding modifications to the user's source code. It only permits writing generated reports to a specific ./codeprobe-reports/ directory, which is a documented and expected behavior for an audit tool.
  • [COMMAND_EXECUTION]: The orchestrator uses the Bash tool to execute local Python scripts (scripts/dependency_mapper.py, scripts/file_stats.py, scripts/complexity_scorer.py) to gather codebase metadata. These scripts utilize only the Python standard library and perform read-only operations on the project directory.
  • [PROMPT_INJECTION]: A low-severity surface for indirect prompt injection is present because the skill ingests arbitrary content from user-provided source files and passes it to an LLM sub-agent.
  • Ingestion points: Source files are read during the 'Pre-Loading Phase' (SKILL.md Section 4) using Glob and Read tools.
  • Boundary markers: The orchestrator wraps file content in '=== FILE: {filepath} ===' and '=== END FILE ===' markers to help the agent distinguish instructions from data.
  • Capability inventory: The sub-agents have access to common agent tools, but are specifically instructed to remain read-only. The orchestrator has the 'Write' tool enabled to save reports.
  • Sanitization: The skill relies on boundary markers and system-level instructions to the sub-agents rather than programmatic sanitization of file content.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 10:55 AM
Security Audit — agent-trust-hub — rightcode-codeprobe