The Agent Skills Directory

[DATA_EXFILTRATION]: The skill exposes local system information by using absolute Windows file paths in the execution_context and process blocks.
Evidence: Files are referenced via C:/Users/nicol/.codex/get-shit-done/..., which leaks the host's username (nicol) and specific directory structure to anyone auditing or receiving the skill output.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) as it ingests untrusted data from user arguments and external files.
Ingestion points: The skill processes user-supplied text via {{GSD_ARGS}} and reads requirements from files specified by the --prd <file> flag.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the ingested data are present in the provided instructions.
Capability inventory: The skill utilizes spawn_agent (via the Task mapping) and request_user_input, which could be manipulated if the input data contains malicious instructions.
Sanitization: There is no evidence of sanitization or validation of the content parsed from the PRD files or arguments before they are used to generate the phase plan.

gsd-plan-phase