Session closing checklist for Git state, documentation, handoff files, and Munin memory. Use when the user invokes /close or asks to wrap up a session, including work that touched multiple repositories.
Resolve a batch of GitHub tickets via parallel headless Claude Code sessions — one isolated git worktree per ticket, gated permissions, triple review (Codex + local-model first-pass + orchestrator comment), interrupt-resume, and merge orchestration. Use when the user wants many well-scoped tickets solved autonomously across one or more repos.
Run an adversarial debate with Claude to stress-test a draft, plan, research claim, or design decision. Use when Codex needs structured cross-model critique before finalizing a consequential decision, architecture, protocol, research claim, security assessment, or product priority.
Run an adversarial debate with a cross-model reviewer (Codex or Antigravity/agy) to stress-test a draft, plan, research claim, or design decision. Use this whenever you need structured critique before finalizing decisions.
Run a cross-model PR review using Codex CLI. Invokes Codex headless to review the current branch diff, then reads findings and fixes issues or reports to user.
Sweep the Munin signals/friction corpus (report_friction self-reports), group recurring failures, and file consolidated GitHub issues routed to their owning repos + the Grimnir Roadmap board. Use when the user wants to triage accumulated friction, turn friction signals into tickets, or asks "what friction have we collected / let's file tickets for it".
Reliable conversation-level email triage over a date range. Identifies genuinely open threads, emails that actually need a reply, and real correspondence hiding in junk — by reconstructing each conversation across inbox + sent + drafts + junk before judging, so mobile-quoted or already-sent replies are never mis-flagged. Use for "what's still open / what do I owe a reply / anything missed" over weeks or months.
Spawn independent AI agents across one or more ecosystems (Claude, Codex, Antigravity) to user-test the Munin memory system. Each explores with minimal pre-knowledge and returns structured feedback on what works, what's broken, and what's missing.