The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Indirect Prompt Injection surface detected. The skill accepts user-provided prompts and sends them to external generative AI models.
Ingestion points: The prompt parameter in FalAPI.generate_image, FalAPI.generate_video, and FalAPI.generate_and_wait in fal_api.py.
Boundary markers: Absent. The prompt is directly interpolated into the JSON payload for the API request.
Capability inventory: Network requests (urllib.request) to the fal.ai API. The skill returns URLs which may be processed by other agent tools.
Sanitization: Absent.
[COMMAND_EXECUTION] (LOW): The skill uses the subprocess module to interface with a local configuration tool.
Evidence: subprocess.run(["clawdbot", "config", "get", f"skill.fal_api.{key}"], ...) in fal_api.py.
Risk: While the command uses a list format to mitigate shell injection and the key name is typically hardcoded, calling external binaries introduces a dependency on the host environment's path and the integrity of the clawdbot binary.
[DATA_EXFILTRATION] (LOW): Network operations are performed to a non-whitelisted domain.
Evidence: self.QUEUE_URL = "https://queue.fal.run" and usage of urllib.request.urlopen in fal_api.py.
Risk: The skill transmits user prompts and potentially source image/audio URLs to the fal.ai infrastructure. While expected for the skill's functionality, it constitutes data transmission to an external service provider.

fal-api