The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill includes instructions to execute inline scripts that modify the underlying configuration of third-party tools (Claude Code and Antigravity) to bypass security-critical trust dialogs.\n
Evidence: SKILL.md contains blocks that write to ~/.gemini/antigravity-cli/settings.json and create custom CLAUDE_CONFIG_DIR environments to set hasTrustDialogAccepted to true automatically.\n- [INDIRECT_PROMPT_INJECTION]: The skill describes a pattern for driving agents where content from external files is interpolated directly into prompts, creating an exploitation surface for indirect prompt injection.\n
Ingestion points: Data is read from brief.md and note.md files.\n
Boundary markers: The skill lacks explicit delimiters or instructions to the sub-agent to ignore potential malicious content within the interpolated files.\n
Capability inventory: Through the herdr CLI, the skill can perform shell operations, split terminal panes, and drive agents with broad file system permissions.\n
Sanitization: The recipes do not include sanitization or validation steps for the content read from files before it is submitted as a prompt.\n- [COMMAND_EXECUTION]: The skill relies on complex shell command sequences, subshell execution ($(...)), and backgrounded processes to manage terminal multiplexers and sub-agents, which increases the risk of command injection.

herdr-faq