code-intelligence
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data as it reads and analyzes source code from the workspace through LSP results and grep output. This represents a potential surface for indirect prompt injection, though it is mitigated by the skill's specific task-oriented focus and the agent's internal safety guardrails.
- Ingestion points: The agent reads source code content at positions returned by LSP operations and text searches (SKILL.md, references/lsp-calls.md).
- Boundary markers: The skill defines a specific 'Disclosure Format' to alert users to tool fallbacks, but does not define explicit delimiters for ingested code data within the scripts.
- Capability inventory: The skill utilizes file system navigation, symbolic search (LSP), and pattern matching (rg).
- Sanitization: No explicit sanitization or escaping of ingested source code is described before interpolation into the agent context.
- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute standard system utilities to maintain environment integrity.
- Evidence: The skill uses
rg(ripgrep) for text search (SKILL.md) and requires verification of tool paths usingtype -a,ls -l, and<tool> --version(references/degradation-and-disclosure.md) to prevent the use of phantom or shimmed binaries.
Audit Metadata