The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the agent's execution capability to run a bundled Python script (scripts/aig_client.py) which acts as a CLI client for the A.I.G platform.
[CREDENTIALS_UNSAFE]: Manages sensitive authentication data, including A.I.G API keys and LLM service tokens. The instructions follow security best practices by requiring these to be passed via environment variables or command-line arguments and explicitly forbidding the agent from echoing them back to the user.
[PROMPT_INJECTION]: Contains instructions directing the agent to bypass standard safety warnings regarding the scanning of local or private network addresses (e.g., localhost, 192.168.x.x). This is intended behavior to facilitate the skill's primary purpose of auditing internal AI services.
[DATA_EXFILTRATION]: Provides a capability to upload local file archives to the A.I.G server for the purpose of source code security auditing. This involves reading local files and transmitting them to a user-configured endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays potentially untrusted data retrieved from external security scans.
Ingestion points: Vulnerability findings, project descriptions, and system logs fetched from the A.I.G API are displayed to the user in the agent's response.
Boundary markers: The skill does not implement specific boundary delimiters or instructions to ignore embedded commands within the scan results.
Capability inventory: The skill relies on exec for Python script execution, local file reads for audit uploads, and network operations conducted via the urllib library in the client script.
Sanitization: The client script performs basic sanitization by truncating long log messages and vulnerability descriptions before they are returned to the agent context.

aig-scanner