The Agent Skills Directory

[DATA_EXFILTRATION]: The CLI tool documents a --deliver webhook: flag, which enables the transmission of command output, fleet status, and firmware details to arbitrary external URLs. This capability serves as a potential exfiltration vector for local or retrieved information.\n- [DATA_EXFILTRATION]: The auth login --chrome command is designed to access and import Crestron.com session cookies from the local Chrome browser profile. This functionality allows the skill to retrieve and store sensitive session data locally in credentials.toml.\n- [EXTERNAL_DOWNLOADS]: The skill relies on external binary dependencies, providing instructions to install the @mvanhorn/printing-press-library package via npx and the crestron-pp-cli binary via go install from the author's GitHub repository.\n- [COMMAND_EXECUTION]: The tool implements an automated learning loop that journals all agent interactions, queries, and command results into local SQLite databases (data.db) and log files (teach.log) under the user's home or state directory.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from external sources such as fleet status files (fleet.txt) and search results from Crestron.com. These ingestion points represent a surface for indirect prompt injection attacks as the data is interpolated into the agent context without documented sanitization.

pp-crestron