demo-guide
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including project source code, configuration files (docs/project-config.json), and documentation (docs/specs/**) to generate its output.
- Ingestion points: Project source code, configuration files, and specification documents referenced during the investigation phase.
- Boundary markers: The skill utilizes specific comment markers (e.g., ) to delimit generated content and repeatedly instructs the agent to use "evidence-based reasoning" and cite file:line references to maintain grounding.
- Capability inventory: The skill leverages various investigation tools ($investigate, $graph-code) and executes a local Python script (.claude/scripts/code_graph) to trace code paths.
- Sanitization: The skill contains mandatory instructions to redact all sensitive values (API keys, tokens, credentials) using redacted:... placeholders if they are encountered during data gathering.
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute a local Python utility (
python .claude/scripts/code_graph) for code analysis. This is a local execution of a project-specific script and does not involve downloading or executing remote code.
Audit Metadata