The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from subagents and external issue trackers, creating a surface for indirect prompt injection. • Ingestion points: The orchestrator reads pull request descriptions, worker-generated questions, relay logs, and issue comments as seen in SKILL.md and escalations-and-provenance.md. • Boundary markers: The instructions emphasize the use of provenance markers such as the 'm:' prefix for maintainer messages and require the signing of all relayed text to distinguish sources. • Capability inventory: The agent can execute rota and herdr CLI commands, manage worktrees, create/merge pull requests, and launch subagents (SKILL.md, solo-and-autopilot.md). • Sanitization: Mitigation is provided through mandatory human-in-the-loop review gates (rota worker gate) and the requirement for the orchestrator to sign and cite the specific channel of all information relayed to workers.
[COMMAND_EXECUTION]: The skill relies on the execution of local CLI tools including rota, herdr, and tmux to perform its core orchestration tasks.
[DYNAMIC_EXECUTION]: In solo mode, the skill involves the dynamic launching of subagents using the Agent tool, passing generated briefs as prompts (solo-and-autopilot.md).

rota-orchestrate