The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from a local configuration file, which is an external ingestion point that could be manipulated by other local processes.
Ingestion points: The agent reads the existing ~/.claude/keybindings.json file.
Boundary markers: There are no specific delimiters or warnings used when reading the JSON content to ignore embedded instructions.
Capability inventory: The skill provides instructions for using file-read, file-write, and edit tools to modify configuration files.
Sanitization: The skill mandates validation of contexts and actions against a whitelist, which partially mitigates risks, but does not explicitly sanitize for natural language instructions.

keybindings-help