The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents and implements multiple ingestion points for untrusted data, including processing file uploads, reading external web URLs, and handling chat history from JSON files. While these are core features for an AI agent, they create a surface where malicious instructions could be embedded in the processed data.
[COMMAND_EXECUTION]: The documentation and reference scripts (e.g., references/tool-builder.md and references/agent-patterns.md) include examples of using eval() to process mathematical expressions from user input. This pattern can lead to arbitrary code execution if the input is not strictly validated or sanitized.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading AI application outputs and fetching remote content via URLs (e.g., https://example.com/skills/api-docs.md or fetching image URLs). These operations are documented as standard platform functionality for processing external resources.
[PRIVILEGE_ESCALATION]: The skill provides patterns for human-in-the-loop workflows to mitigate risks associated with sensitive operations like file deletion, though the underlying capability to perform such operations exists within the agent's potential toolset.

python-sdk