The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configures persistence on the host operating system by creating launchd plists in ~/Library/LaunchAgents/ (macOS) or modifying crontab (Linux) to execute scheduled tasks via the openpaw command-line tool.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because its primary use cases involve processing untrusted content from external sources like emails and GitHub notifications.
Ingestion points: The skill reads and summarizes untrusted external data from email systems and GitHub activity.
Boundary markers: No specific delimiters or boundary markers are identified in the instructions to protect the model from embedded malicious commands in processed data.
Capability inventory: The skill possesses capabilities to write to the local filesystem (specifically ~/.config/openpaw/), execute system-level commands through schedulers, and transmit data externally via the Telegram delivery method.
Sanitization: There is no documented evidence of input validation or sanitization for the external data being summarized.
[DATA_EXFILTRATION]: The skill maintains sensitive data including task results, cost logs, and job schedules in the user's home directory (~/.config/openpaw/). It also supports a telegram delivery method which transmits task results to an external network endpoint.

c-schedule