The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses sensitive local files to verify configuration and installation status.
The cursor-install.md script reads Cursor's internal SQLite state database (state.vscdb) to check for administrative policy flags (allowUserLocalPluginImports, allowThirdPartyPluginImports).
The skill reads and appends data to system shell configuration files (~/.zshrc, ~/.bashrc) to manage environment variables.
[PERSISTENCE]: The skill implements a persistence mechanism to ensure the CLI forwarder is available in future sessions.
It appends an export PATH command to the user's shell profile files (.zshrc and .bashrc).
[DYNAMIC_EXECUTION]: The skill generates and executes local script content at runtime.
It creates a shell forwarder script at ~/.local/bin/clay using a heredoc and grants it execution permissions (chmod +x).
It utilizes inline Python scripts (python3 -) to perform complex JSON and SQLite processing on local state files.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files and tool outputs which could potentially influence agent behavior.
Ingestion points: Reads admin policy settings from Cursor's state.vscdb database, parses Cursor Plugins.log for status strings, and consumes JSON output from the clay whoami CLI command.
Boundary markers: None identified; data is processed directly via shell and Python logic.
Capability inventory: Includes the ability to write files (cat, printf), remove directories (rm -rf), and execute shell commands (exec).
Sanitization: Employs specific SQL queries and JSON parsing (json.loads) to safely extract targeted configuration values from complex data structures.

setup