The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bootstrap script (scripts/run.py) to manage a virtual environment and execute its CLI. It uses list-based subprocess.run calls, which prevents shell-based command injection by avoiding direct shell interpolation.
[CREDENTIALS_UNSAFE]: Authentication is handled via the SENSORTOWER_AUTH_TOKEN environment variable. The skill includes explicit redaction logic in scripts/sensortower.py to strip the token from API error responses and diagnostics, preventing accidental exposure.
[EXTERNAL_DOWNLOADS]: The runner script is designed to install future dependencies from requirements.txt using pip install --require-hashes. This ensures that only cryptographically verified packages can be introduced into the skill's execution environment.
[DATA_EXFILTRATION]: Network activity is restricted to the Sensor Tower API. The code validates the SENSORTOWER_API_BASE_URL environment variable to ensure it is a valid HTTPS origin, preventing protocol smuggling or redirection to unsafe endpoints.
[PROMPT_INJECTION]: The skill processes data from a third-party API and instructs the agent to summarize it. While it lacks explicit boundary markers in the prompt instructions to delimit this external data, the use of structured JSON and sanitized error reporting reduces the risk of indirect prompt injection. Ingestion point: request_json in scripts/sensortower.py. Boundary markers: Absent. Capability inventory: Managed subprocess and HTTPS requests. Sanitization: JSON parsing and specific redaction logic.

sensortower