The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill contains a shell script that searches for a specific file (gsd-tools.cjs) across numerous potential installation directories (including ~/.claude, ~/.hermes, ~/.cursor, and others) and executes it using node. This runtime discovery and execution of local scripts at computed paths is a dynamic execution pattern.
Evidence: Found in the <process> block of SKILL.md.
[COMMAND_EXECUTION]: The instructions direct the agent to append user-supplied arguments (such as a directory path) to a shell command (gsd_run from-gsd2). This creates a risk of command injection if the agent fails to properly sanitize the user-provided input before execution.
Evidence: The instruction "append --path <dir> if provided" in SKILL.md combined with the shell execution block.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external plan files via the --from argument. This is an untrusted data ingestion point where malicious instructions could be embedded to influence the agent.
Ingestion points: The --from <filepath> argument allows reading external files into the agent's context.
Boundary markers: None present; the instructions do not specify using delimiters or warnings for the imported content.
Capability inventory: The skill has access to Bash, Write, Edit, and Agent tools, which could be abused if an injection is successful.
Sanitization: No sanitization or validation of the external file content is mentioned.

gsd-import