The Agent Skills Directory

[COMMAND_EXECUTION]: Employs curl to interact with Ironclad's API for listing and creating workflows, records, and webhooks. These commands utilize environment variables for instance configuration and authentication.
[EXTERNAL_DOWNLOADS]: Fetches contract documents from the configured Ironclad host and saves them to the local filesystem (/tmp/contract.pdf). While automated scans flagged this as a potential remote code execution risk, the skill documentation only covers the download operation without instructions to execute the resulting file.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8) by ingesting untrusted data from an external platform (Ironclad).
Ingestion points: Data retrieved from the Ironclad API host ($IRONCLAD_HOST) in response to GET requests for workflows and records (SKILL.md).
Boundary markers: None identified. The skill does not instruct the agent to treat data from the API as untrusted or separate it from system instructions.
Capability inventory: File system write access (/tmp), network operations (curl), and document downloading (SKILL.md).
Sanitization: No evidence of sanitization, validation, or escaping of the content returned by the API before processing.
[DATA_EXFILTRATION]: Sensitive contract metadata and POST request bodies are written to temporary files in /tmp (e.g., /tmp/ironclad_workflow.json). This practice could lead to local data exposure in shared or insecure compute environments.

ironclad