The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/launch.py and scripts/collect.py) that manage worker agents by invoking the herdr CLI tool.
Subprocess calls in the launcher script use list-based arguments rather than shell strings, effectively preventing command injection at the script level.
The launcher validates worker configurations against a strict schema and a whitelist of allowed CLI types (pi, codex, claude), rejecting any identifiers containing control characters.
The launch.py script implements a monitor service that uses a 32-character secret token for local socket authentication to ensure only the authorized collector can access run events.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to synthesize reports from multiple worker agents who process external project data, creating a potential surface for indirect injection.
Ingestion points: The skill resolves user-provided requests/scopes in SKILL.md and aggregates reports from worker agents via scripts/collect.py.
Boundary markers: The lead agent prepares a common 'brief' for all workers, which is passed to the launcher and delivered to worker agents using quoted heredocs (<<'QUORUM_BRIEF') to prevent unintended shell expansion of the brief's content.
Capability inventory: The skill's scripts facilitate terminal pane management and agent orchestration (starting, prompting, and interrupting agents) via the herdr tool.
Sanitization: In addition to validating worker configuration tuples, the launcher uses json.dumps for specific arguments to ensure they are correctly escaped before being passed to the worker agents.

herdr-quorum