The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to construct and execute shell commands (uv run) using user-provided parameters such as prompt and filename. Although the instructions advise the agent to perform safe escaping, building shell commands through string interpolation of untrusted data is a classic command injection surface.
[REMOTE_CODE_EXECUTION]: The skill depends on a script located at an external path (~/.agents/skills/openrouter-img/scripts/generate_image.py) which is not included in the skill's source files. This represents an unverifiable dependency executing code from a location outside the immediate project scope.
[DATA_EXFILTRATION]: The skill explicitly directs the agent to read sensitive data (OPENROUTER_API_KEY) from .env or ~/.env. Accessing hidden environment files, particularly in the user's home directory, constitutes sensitive data exposure.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User-supplied image prompts, filenames, and aspect ratios processed by the agent (SKILL.md).
Boundary markers: None identified; user input is interpolated directly into command arguments.
Capability inventory: Subprocess execution via uv run (SKILL.md).
Sanitization: The skill includes a natural language instruction to "perform safe escaping," but lacks programmatic validation or strict schema enforcement for the inputs.

to-img