The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data extracted from PDF and image invoices. A malicious document could contain hidden instructions intended to override agent behavior.
Ingestion points: Invoice file content (PDF text, image OCR) and original filenames extracted during the scanning process in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings for the agent to treat extracted document text strictly as data rather than instructions.
Capability inventory: The agent has access to file system operations including finding files (find), creating directories (mkdir), and copying/moving files (cp, mv) as described in the execution steps.
Sanitization: The skill includes a step to sanitize proposed filenames by removing special characters, which helps mitigate basic command injection, but there is no explicit sanitization or escaping for the data extracted from within the documents.
[COMMAND_EXECUTION]: The skill instructs the agent to execute local shell commands to perform its core tasks.
Evidence: Uses find to discover files, mkdir -p to build folder structures, and cp or mv to organize documents into the new hierarchy.
Context: These operations are necessary for the skill's stated purpose of invoice organization and are restricted to the local file system based on user-provided directories.

invoice-organizer