The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The scripts scripts/agnes-image.py and scripts/agnes-video.py access local files at ~/.pi/agent/auth.json and ~/.pi/agent/models.json. These files typically contain sensitive authentication tokens and provider credentials used by the agent platform.
[COMMAND_EXECUTION]: Both Python scripts implement a _resolve_value function that uses subprocess.run(shell=True) to execute strings prefixed with ! found within the agent's configuration files. This allows for arbitrary shell command execution during the API key resolution process.
[DATA_EXFILTRATION]: The skill's image-to-image and image-to-video features read local files and transmit their contents as Base64-encoded strings to the external API endpoint https://apihub.agnes-ai.com/v1/images/generations. While this is functional for the skill, it involves sending local data to a remote service.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted inputs (prompts and remote URLs) that could contain malicious instructions for the AI agent.
Ingestion points: The prompt, --input-image, and --keyframes arguments in scripts/agnes-image.py and scripts/agnes-video.py accept external content.
Boundary markers: The scripts do not use delimiters or instructions to ignore embedded commands when passing these inputs to the remote API.
Capability inventory: The skill possesses capabilities for network access, local file reading, and shell command execution via its configuration resolution logic.
Sanitization: There is no evidence of input validation or escaping for the user-provided prompt strings or URL parameters.

create-image-video