The Agent Skills Directory

[COMMAND_EXECUTION]: The skill installs a persistence mechanism by copying a property list file to ~/Library/LaunchAgents/ and using launchctl load to ensure a cleanup script runs automatically on the system.
[COMMAND_EXECUTION]: Multiple shell commands are used to install software (brew install screenpipe), modify user configuration files (~/.screenpipe/config.json), and manage local processes (pkill screenpipe, screenpipe &).
[DATA_EXFILTRATION]: The skill enables continuous screen capture and OCR. While it implements a blocklist for browsers and password managers, there is an inherent risk of capturing sensitive PII or credentials if they are displayed in permitted applications like Slack, Notion, or terminals.
[PROMPT_INJECTION]: The skill introduces an indirect prompt injection surface by processing untrusted data (OCR text from the screen) to answer user queries.
Ingestion points: Captured OCR text from screenpipe is ingested into the agent context for summarization and retrieval tasks.
Boundary markers: None provided in the instructions to separate retrieved OCR data from system instructions.
Capability inventory: The agent has the ability to execute shell commands, install software, and read/write local configuration files.
Sanitization: No sanitization or verification of the OCR content is performed before processing.

screenpipe-setup