The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to interact with external systems.
Uses curl to perform network requests to target SharePoint endpoints, including GET and POST operations with custom headers and SOAP bodies.
Uses jq and grep to parse and extract data from remote server responses.
Uses python3 for local string processing.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external servers, creating a vulnerability surface where the agent's behavior could be influenced by malicious content.
Ingestion points: The agent reads and processes data from various file paths (SKILL.md), including HTTP headers (e.g., SPRequestGuid, X-SharePointHealthScore), HTML bodies (e.g., __REQUESTDIGEST, LibraryVersion), and JSON API responses.
Boundary markers: The instructions do not include explicit boundary markers or warnings to the agent to disregard instructions embedded in the data retrieved from external servers.
Capability inventory: The agent has capabilities to perform network operations (curl) and execute local code (python3), which could be exploited if the agent is compromised by malicious input from a target server.
Sanitization: While the skill uses urllib.parse.quote to sanitize class names before interpolation into URLs, other data points extracted from remote responses are processed directly.
[DYNAMIC_EXECUTION]: The skill utilizes runtime script generation to perform specific utility tasks.
Uses python3 -c to dynamically generate and execute code for URL-encoding class names for reflection-based enumeration.
References the use of a Python snippet for decoding NTLM protocol data (AV-pairs) captured from server responses.

hunt-sharepoint