The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes data from external financial API interfaces (via the pandadata-api dependency) and interpolates it into the agent's prompt without explicit boundary markers or instructions to ignore embedded malicious instructions.\n- Ingestion points: Market data retrieved from pandadata-api and user-supplied strings in the symbol and task fields within the qsh-form template.\n- Boundary markers: The prompt template lacks explicit delimiters or instructions to the agent to treat the external data as untrusted content, relying instead on a fixed 9-chapter reporting structure.\n- Capability inventory: The skill utilizes shell execution for data retrieval (referencing a local scripts/call_api.py) and indirect network operations through the Pandadata API.\n- Sanitization: There is no evidence of sanitization, filtering, or escaping for user-provided input or API responses before they are included in the prompt template.

a-share-stock-dossier