A single-purpose orchestration primitive that assembles review panels based on file types and keywords, then hands back a structured recommendation to whatever called it. You don't invoke this directly. It's machinery for /review, /plan-review, and code-reviewer to decide who should look at a diff or plan file. The policy algebra is frozen at version 1, which is both disciplined and slightly ominous. Technical-editor is always included, security-auditor gets pulled in for anything touching secrets or workflows, and there's a hard cap of 3 rounds before escalation. The orchestrator authority section is unusually explicit about when you can push back on findings and when you absolutely cannot, which suggests this emerged from real friction in a multi-agent review loop.
npx skills add https://github.com/camacho/ai-skills --skill assemble-panel