The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and processing data from other AI agents (read-backs, reports) and external tools (Git commit IDs).
Ingestion points: Data enters the orchestrator's context via agent read-backs in the READ BACK section and status checks in the Score section of SKILL.md.
Boundary markers: The huddle protocol uses explicit headers (SITUATION, PLAY, SNAP, etc.) which act as delimiters, though they are not robust against adversarial content in the read-backs.
Capability inventory: The orchestrator utilizes tools for agent communication (SendMessage, codex queue), version control (git), and a shell script (wait-all-idle.sh) for synchronization.
Sanitization: The process relies on the LLM to manually verify that agent read-backs match the requested 'play', but no automated sanitization or escaping of external content is specified.
[COMMAND_EXECUTION]: The skill instructions specify the use of shell-based tools and version control systems.
Evidence: Instructions in SKILL.md direct the agent to use git for status checks and to execute wait-all-idle.sh with specific arguments (herdr agent wait <pane> --until working) to manage the multi-agent workflow.

huddle