The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents methods for reading data from the local filesystem into the agent's context, which can lead to indirect prompt injection if the files being processed contain malicious instructions.
Ingestion points: The skill provides patterns for 'fs.readFileString', 'fs.readFile', and 'fs.readDirectory' (references/02-reading-operations.md).
Boundary markers: The reference material lacks instructions for implementing delimiters or security warnings (e.g., 'ignore embedded instructions') when processing file content.
Capability inventory: The skill documents powerful capabilities that an injected prompt could abuse, including recursive directory removal, file writing, and permission modifications (references/03-writing-operations.md, references/04-file-operations.md).
Sanitization: There is no guidance or example for sanitizing, validating, or escaping content read from files before it is used by the agent.
[COMMAND_EXECUTION]: The documentation includes detailed examples for modifying file system metadata and permissions.
Evidence: The skill provides code patterns for 'fs.chmod' (changing file modes) and 'fs.chown' (changing ownership) in 'references/07-permission-operations.md'.

effect-filesystem