The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of documents (PDF, Word, Images, etc.), which creates an attack surface for indirect prompt injection.
Ingestion points: The somark_parser.py script reads local file content or user-provided files for processing.
Boundary markers: The SKILL.md instructions include explicit safety rules directing the agent to treat all parsed output strictly as data and to ignore any embedded role instructions, jailbreak attempts, or override requests.
Capability inventory: The skill possesses file read/write capabilities and performs network operations to transmit document data to an external API.
Sanitization: The skill relies on the agent's adherence to instructional boundaries rather than programmatic sanitization of the parsed document content.
[EXTERNAL_DOWNLOADS]: The skill transmits document content to external servers for parsing.
Evidence: The script somark_parser.py communicates with https://somark.cn/api/v1 and https://somark.ai/api/v1. These are the official endpoints for the SoMark service and align with the skill's stated purpose.
[CREDENTIALS_SAFE]: The skill demonstrates best practices for secret management.
Evidence: The instructions explicitly warn the user not to share their API key in the chat and instead direct them to configure the SOMARK_API_KEY environment variable, which the Python script then retrieves safely using os.environ.get().

somark-document-parser