seed-test-data
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill operates by ingesting and processing content from local documentation files (
docs/project-config.json,docs/project-reference/*,lessons.md, etc.) and the output of search commands to determine its logic. If these repository files are modified by an external actor, they could influence the agent's behavior during code generation. - Ingestion points: Reads local configuration and project-reference documents via direct file access and filesystem searches.
- Boundary markers: The instructions require the agent to cite evidence with
file:linereferences, which provides a level of traceability. - Capability inventory: The skill utilizes filesystem searching (
rg), sub-agent spawning (spawn_agent), and execution of local scripts. - Sanitization: There is no explicit sanitization or filtering logic described for the content read from the documentation files before it is processed.
- [COMMAND_EXECUTION]: The skill makes extensive use of the
rg(ripgrep) utility to search for code patterns, base classes, and configuration keys within the local source code. - [REMOTE_CODE_EXECUTION]: The skill is designed to execute a local Python script (
.claude/scripts/code_graph) to perform dependency tracing and code analysis. This involves running code that is bundled within the user's repository.
Audit Metadata