The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted email content from Gmail, which is a vector for indirect prompt injection.
Ingestion points: SKILL.md defines search and read functions (+triage, +read) as data entry points.
Boundary markers: Instructions mandate that agents "Treat search results as candidate evidence, not fact" and require human-in-the-loop (HITL) confirmation for all write operations.
Capability inventory: The skill can perform actions like +send, +trash, and +archive as listed in SKILL.md.
Sanitization: The skill advises on shell-quoting guards for dynamic data to prevent CLI command injection.
[DYNAMIC_EXECUTION]: The skill requires the agent to manage local files and decode data at runtime.
Evidence: SKILL.md instructs the agent to decode base64 attachment data and write email content to files for the --upload operation.
Context: These actions are core to the skill's functionality and are conducted through the platform's provided CLI tool.

gmail