The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches version metadata and the official changelog from Anthropic's public GitHub repository to ensure runtime compatibility and feature availability.
[COMMAND_EXECUTION]: To support its primary purpose of orchestration, the skill dynamically generates and executes management commands for various sub-agent backends including gc, Codex CLI, and Claude Native Teams.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting wave data from external JSON files (via the --from-wave flag).
Ingestion points: scripts/ol-wave-loader.sh parses and validates wave entries from local JSON files.
Boundary markers: Data from the JSON file (id, title, spec_path) is interpolated into task descriptions without explicit boundary delimiters.
Capability inventory: The skill utilizes powerful tools such as Bash, Task, TeamCreate, and SendMessage to execute its workflow.
Sanitization: Input validation is performed in scripts/ol-wave-loader.sh to prevent command injection via control characters (newlines/tabs) and uses jq for structured parsing.

swarm