The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted external data from various file formats (DOCX, XLSX, PDF, and images). This content is extracted and presented to the agent's context for verification, which could allow maliciously crafted documents to influence agent behavior.
Ingestion points: The scripts/verify.py tool reads content from files using libraries like python-docx, openpyxl, fitz, and RapidOCR.
Boundary markers: The skill lacks explicit boundary markers or "ignore instructions" warnings when presenting extracted text to the agent.
Capability inventory: The skill has the capability to execute shell commands (e.g., pytest, ruff) and perform file system operations (creating new files, deleting verified files).
Sanitization: There is no evidence of sanitization or filtering of the text extracted from these external documents before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructions in rules/checklists.md direct the agent to execute shell commands such as pytest and ruff. While these are standard development tools, they represent a vector for executing code within the project's environment as part of the verification process.

verify-deliver