The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a set of local Bash scripts (e.g., herdr-launch, herdr-worker, herdr-say) to interface with the herdr CLI tool for managing workspaces, worktrees, and agent processes. These scripts are functional wrappers for the skill's orchestration capabilities.
[INDIRECT_PROMPT_INJECTION]: The multi-agent workflow creates a surface for indirect prompt injection, where data or instructions from one agent role could potentially influence the behavior of another in the chain.
Ingestion points: Workers ingest tasks from the Admin; the Admin ingests status reports and sentinels from Workers; the Main agent ingests milestone reports from the Admin.
Boundary markers: The skill defines specific sentinels (e.g., 'WORKER DONE', 'NEEDS-INPUT') to structure and delimit communication between agent roles.
Capability inventory: Orchestration scripts invoke the herdr CLI to perform repository operations, process management, and communication across panes.
Sanitization: Role instructions in roles/worker.md explicitly mandate that agents validate all external input before use.

herdr-orchestration