The Agent Skills Directory

[COMMAND_EXECUTION]: The skill heavily utilizes the herdr CLI to launch subagents, send keystrokes to terminal panes, and manage lifecycle states. This includes complex shell piping and orchestration logic to drive external processes.
[DATA_EXFILTRATION]: Accesses and modifies sensitive local configuration files, specifically ~/.gemini/antigravity-cli/settings.json and .claude.json. These actions are performed to programmatically bypass folder-trust security dialogs intended to protect the user's filesystem from unauthorized agent access.
[DYNAMIC_EXECUTION]: Employs a Python heredoc to dynamically generate and execute a script at runtime that modifies JSON configuration files to inject trusted workspaces.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads raw terminal output and detection state from third-party agents using herdr agent read and herdr pane read (SKILL.md).
Boundary markers: No explicit delimiters or boundary markers are used when processing agent output; logic relies on raw text matching via grep and jq.
Capability inventory: The skill has capabilities to start/rename agents, send prompts, and modify local configuration files.
Sanitization: There is no evidence of sanitization or escaping of the ingested agent output before it is used to influence the skill's decision logic or troubleshooting steps.

herdr-faq