Operating with agents
Yoke attaches to agent hosts (Claude Code, Codex, Cursor, …). The harness loads skills and hooks; Yoke owns state, approvals, and evidence.
Session loop
/yoke steer # staff work from strategy
/yoke charge # run frontier head
/yoke feed # refresh frontier / materialize from strategy
/yoke strategize # guided strategy review
Delivery adapters
/yoke idea | dash | blitz | shepherd | conduct | implement | polish | usher
Exact stage ownership comes from the item's pinned workflow version — read
yoke workflows item get PREFIX-N, do not memorize progressions.
Rules of thumb
- One work item → one project deploy target
- Implementation happens in worktree lanes, not on main
- Claims and Doctor findings are coordination facts — fix or coordinate, do not descope required files to dodge them
- Prefer registered
yokecommands and function ids over ad hoc scripts
Command map: Commands.
Staff work through /yoke steer; inspect the runnable frontier with /yoke charge.
Related
- Commands
Find operator commands and the delivery skill for the current binding.
- Maintenance commands
Inspect maintenance operations, internal helpers and project context.
- Native discovery
Load shared instructions, discover skills and inspect installed capabilities.
- Codex
Install and verify Yoke’s Codex hooks and agent surfaces.
- Cursor
Install and verify Yoke’s Cursor hooks and agent surfaces.
- Execution levels
Read model thresholds, routing settings and stage-level handoffs.
- Level launches
Launch and label sessions using model thresholds and native selectors.
- Capacity recovery
Recover a launched session when its model capacity is exhausted.
- Operation homes
Choose the rule home for the operation before acting.