The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external, untrusted contract files (txt, md, pdf) which creates a surface for indirect prompt injection. Malicious instructions embedded in a contract could attempt to manipulate the extraction logic or the subsequent update to config.json.
Ingestion points: The command python3 scripts/contract.py read <file> reads external text files (SKILL.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are mentioned.
Capability inventory: Subprocess execution of scripts/contract.py, filesystem writes to .fukugyo/contracts/, and modification of config.json (SKILL.md).
Sanitization: No sanitization or validation of the input text is described before LLM processing.
[COMMAND_EXECUTION]: The skill requires the execution of a local Python script (scripts/contract.py) to parse documents and manage data. Users should verify the script content as it is not provided in the skill definition.
[DATA_EXFILTRATION]: The skill is designed to access and structure sensitive business information from contracts, such as compensation rates and payment terms. This data is stored locally in a hidden directory (.fukugyo/contracts/).

fukugyo-contract