The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection as it retrieves and processes data from external systems. Ingestion points: Untrusted content enters the agent context through tools like mcp__glean__company_search, atl jira issue view, atl confluence page view, and gh pr view. Boundary markers: The instructions do not define delimiters or specific 'ignore embedded instructions' warnings for the data being processed. Capability inventory: The agent is authorized to execute commands via the atl, gh, and git CLIs, and has access to the local file system. Sanitization: No validation or sanitization of the retrieved content is mentioned in the research methodologies.
[COMMAND_EXECUTION]: The skill relies on several CLI tools to perform its functions, including atl (for Jira/Confluence), gh (for GitHub), and standard git. It also suggests the use of rails runner and rails console for interacting with development environments in specific troubleshooting scenarios.
[DATA_EXFILTRATION]: The skill facilitates access to sensitive internal company information, including architectural decision records (ADRs), RFCs, and repository history. These operations are performed using legitimate tools targeting official ezCater domains.

ezcater-research