Overview
dsh-tier-router
Tiered model routing for DeepSeek Harness (+ fallback chains + task-intensity effort): a strong tier for planning/architecture/review, a cheap tier for implementation, with plan-mode-aware auto routing, per-tier model fallback chains, task-intensity reasoning effort selection, advisor/review consultations, a high-impact escalation guard, failure auto-escalation, durable tier configuration, and subagent tiering.
README / EN
Package documentation
Registry summary
Tiered model routing for DeepSeek Harness (+ fallback chains + task-intensity effort): a strong tier for planning/architecture/review, a cheap tier for implementation, with plan-mode-aware auto routing, per-tier model fallback chains, task-intensity reasoning effort selection, advisor/review consultations, a high-impact escalation guard, failure auto-escalation, durable tier configuration, and subagent tiering.
dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.
Read the full README on GitHubLIMITATIONS
Known limitations
- Tier switches take effect from the next step (the header is written before the step builds). - The subagent policy (`/tier subagent`) is process-wide and classifies child execution for the guard. A worker's tier is fixed at creation (`tier_worker` agentOptions); built-in `subagent`/`subagent_fork` children inherit the parent's model route. - The full live "fail -> self-heal" sequence for failure auto-escalation has been verified at the event path and unit-logic level; trigger it live per the FAQ if desired.