Spec-first long-horizon execution. Use for work the chair would otherwise avoid as 'too big for one shot' — a whole subsystem, a greenfield feature, an end-to-end migration, a sweep across dozens of files. Forces a complete written SPEC to disk BEFORE any…
Work on anything that talks to herdr — the herdr-extensions package, the herdr-edit editor fork (the "tiny VS Code for herdr" stack), or a herdr API client such as the Herdr Fleet view in vzt-agent-protocol/vscode. Loads the invariants that fail SILENTLY…
Visual design lane — author a repo's DESIGN.md from its real token layer, then apply it. Use for restyling, rethemes, brand passes, spacing/typography/palette work, 'make this look right'. Not for technical architecture (that is /vzt-design). Usage: /vzt-ui…
VZT Agent Protocol routing doctrine and manual router. Use when the user asks how work is being routed, wants to route a specific task ('/vzt-route <task>'), wants routing stats, or when deciding which model tier should handle a piece of work.
Plan the given task on Opus 5 at max effort — the `opus@max` rung. Use for architecture, technical specs, roadmaps, migration plans, and breaking down a feature when the planning needs full conversation context. This is the DEFAULT planning turn; use…
Frontier working discipline as a portable process — the five gates (scope, evidence, attack, verify, report) extracted from the Fable-tier agents so ANY model tier can run them. Use when executing delegated work, when a cheaper tier is handling up-tier-shaped…
Plan the given task on Fable 5 with full conversation context — the LAST rung. Use ONLY when the design has no prior art to reason from: novel or greenfield architecture, from-scratch system design, one-way-door sharding/replication/consensus/multi-tenancy…
Parallel hypothesis fan-out for a hard bug — enumerate N candidate root causes, dispatch N read-only agents to confirm-or-refute each one with a real command, and collect evidence before spending a frontier tier. Use when a bug has 2+ plausible causes, BEFORE…