The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration workflow in SKILL.md defines shell command templates that interpolate user-provided parameters like scope_subpath and output_dir. There are no instructions for the agent to sanitize these inputs, creating a surface for potential command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Go repositories which can contain malicious instructions intended to manipulate the worker agents. Ingestion points: Go source files, module definitions, and inventory data processed throughout the orchestration workflow. Boundary markers: The workflow does not define specific delimiters or instructions for workers to ignore embedded commands within the analyzed code context. Capability inventory: The skill executes multiple shell scripts via python3 and manages worker agents that perform the analysis. Sanitization: There is no evidence of sanitization or filtering of the analyzed repository content before it is processed by the agents.

go-review