The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches source code and script assets from an external repository at https://github.com/beatra-ai/ai-image-realism-skill.git using git clone and checkout commands.
[REMOTE_CODE_EXECUTION]: Executes downloaded Python scripts, including authorize.py, mcp_client.py, and uninstall.py, which perform network operations and local file modifications.
[COMMAND_EXECUTION]: Employs shell commands including git, cp, and python3 to manage the installation, verification, and operation of the skill.
[CREDENTIALS_UNSAFE]: Accesses and stores a bearer token in the sensitive local path ~/.beatra/credentials.json, which is shared across all skills from the same vendor on the device.
[DATA_EXFILTRATION]: Transmits system telemetry including hostname, OS version, CPU architecture, and Python version to beatra.ai and beatra.cn. It also routes user-supplied media and prompts through infrastructure in mainland China when the primary network is unavailable or latent.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied images and prompts which are sent to external third-party AI models. Ingestion points: User-supplied image files and descriptive prompts. Boundary markers: Absent. Capability inventory: Subprocess execution (python3) and network operations (HTTPS). Sanitization: No explicit sanitization or filtering of external content is described.
[TIME_DELAYED_CONDITIONAL]: Implements a version-checking mechanism that connects to a remote JSON file (sources.json) every 24 hours to check for updates.

ai-image-realism