The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and relays content from multiple external agents and machines, creating a vulnerability to indirect prompt injection from the agents being monitored.
Ingestion points: The skill reads status, agent lists, and transcripts from local and remote Herdr instances using herdr agent list, herdr machine list, and herdr pane read (SKILL.md).
Boundary markers: Instructions require the agent to use verbatim quotes for all relayed messages and to clearly identify the source orchestrator for every relay to maintain attribution (SKILL.md).
Capability inventory: The skill uses the herdr CLI for monitoring, background waiting, and UI focus control, and executes a local helper script herdr-agent-prompt-if-input-empty.sh for message transport (SKILL.md).
Sanitization: No explicit sanitization or filtering of external input is performed; the skill relies on exact transcription and user oversight for safety.
[COMMAND_EXECUTION]: The skill relies on the herdr CLI tool for primary operations, including listing agents, checking machine status, and managing workspace focus. It also invokes a specific bash script herdr-agent-prompt-if-input-empty.sh to prevent overwriting active user input during message relays.
[EXTERNAL_DOWNLOADS]: The skill depends on a script located at ~/.agents/skills/shunk031-herdr-orchestrate-workers/scripts/herdr-agent-prompt-if-input-empty.sh. This is a dependency on a resource from the same vendor to provide operational safeguards during communication.

shunk031-herdr-meta-orchestrate