The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of external code from the vendor's infrastructure:
Downloads the @mvanhorn/printing-press-library from the npm registry.
Installs the sensortower-pp-cli directly from the vendor's GitHub repository (github.com/mvanhorn/printing-press-library).
[COMMAND_EXECUTION]: The skill operates by executing shell commands, including software installation via npx and go install, and running the sensortower-pp-cli binary. It also suggests backgrounding commands with the & operator (e.g., sensortower-pp-cli teach ... &).
[CREDENTIALS_UNSAFE]: The skill features an authentication command (auth login --chrome) designed to import session cookies from the user's Chrome browser, which constitutes a sensitive credential harvesting operation.
[DATA_EXFILTRATION]: The tool includes an output delivery feature (--deliver webhook:<url>) capable of sending command results, which may contain sensitive app rankings or metadata, to an arbitrary external URL.
[PROMPT_INJECTION]: The skill uses a 'learning loop' (recall/teach) that creates an indirect prompt injection surface:
Ingestion points: Untrusted data enters the agent context through 'learnings' and 'playbooks' stored in a local SQLite database (data.db).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when retrieving stored learnings or replaying playbooks.
Capability inventory: The skill can perform network operations, file system writes, and shell command execution via the CLI.
Sanitization: There is no automated sanitization of the queries or notes stored in the learning loop, relying instead on a manual 'PII rule' for the user to follow.

pp-sensortower