The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by ingesting untrusted external data.
Ingestion points: The instructions.md file specifies that the skill accepts data in common formats including CSV, Excel, and JSON, and processes user-provided project data or file paths.
Boundary markers: There are no explicit boundary markers or delimiters defined to separate user-provided data from system instructions, nor are there specific instructions for the agent to ignore embedded commands within the processed data.
Capability inventory: The skill utilizes pandas for data processing and has filesystem permissions as defined in claw.json. While the provided Python implementation in SKILL.md is limited to calculation logic, the combined access to the file system and the processing of structured data files creates an injection surface.
Sanitization: The instructions.md file contains a general instruction to "Validate inputs before processing," but no specific sanitization, schema validation, or adversarial filtering logic is implemented in the provided code.

labor-rate