The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the Instantly API and documentation, which could theoretically contain malicious instructions.
Ingestion points: Uses the WebFetch tool to retrieve documentation, OpenAPI schemas, and package metadata from developer.instantly.ai and api.instantly.ai. It also processes tool output from Read, Glob, and Grep when inspecting local files.
Boundary markers: The skill includes explicit instructions to "Redact API keys, addresses, lead payloads, reply content, and email bodies" and specifies that mutations require "explicit approval from the accountable owner."
Capability inventory: The skill is granted Write and Edit permissions, allowing it to modify files based on processed data.
Sanitization: Instructions mandate the redaction of sensitive fields and measured alerting against baselines, though specific prompt-level delimiters for external content are not explicitly defined.

instantly-observability