The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local shell commands (curl) for voice notifications across all workflows (e.g., SKILL.md, TestIdea.md). These commands target a service running on localhost:8888. While intended for user feedback, executing shell commands with interpolated content can be a security risk if the content is not properly sanitized.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in the TestIdea.md workflow. It processes user-supplied ideas and research data, then interpolates the resulting summary into a shell command.
Ingestion points: User-provided ideas and concepts are ingested in Workflows/TestIdea.md, and external research data is ingested via the Research skill in Workflows/UpdateModels.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the user idea or research data as untrusted content that should not be obeyed as instructions.
Capability inventory: The skill uses curl command execution in every workflow file (TestIdea.md, UpdateModels.md, ViewModels.md).
Sanitization: There is no evidence of sanitization or escaping of the {SUMMARY_OF_EXECUTIVE_VERDICT} or other variables before they are placed into the curl data payload.

WorldThreatModelHarness