The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation process involves cloning an external repository from GitHub (beatra-ai/ai-audiobook-generator-skill) and executing the bundled Python scripts (authorize.py, mcp_client.py). This involves running code from a third-party source.
[DATA_EXFILTRATION]: The skill transmits system information (hostname, OS version, CPU architecture) as telemetry to external domains. It also uploads user-supplied manuscript files to synthesis servers (mcp.beatra.ai and mcp.beatra.cn).
[CREDENTIALS_UNSAFE]: The authorization script generates and stores a bearer token with spending capabilities (wallet:spend) in ~/.beatra/credentials.json. The token is shared across all skills for the service on the device.
[COMMAND_EXECUTION]: The documentation provides instructions for the agent to execute shell commands for installation (git clone), verification (grep, awk), and operation (python3), which involve interacting with the network and local filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process manuscripts and course texts.
Ingestion points: Untrusted data enters via manuscripts or course texts provided by the user (as described in the 'When to Use This Skill' section).
Boundary markers: The provided documentation does not specify the use of delimiters or 'ignore' warnings for the ingested data.
Capability inventory: The skill uses subprocess calls to git, python3, and performs network PUT operations for file uploads.
Sanitization: No evidence of input sanitization or validation of the manuscript content is provided.

ai-audiobook-narration