Two-tier model router for DeepSeek Harness
green-dalii/dsh-shift-router · model-adapters
Two-tier model router for DeepSeek Harness — LLM-Judge routing, multi-model fallback chains, exponential-backoff failover, and task-level orchestration (DSH adaptation of pi-shift-router).
★ 2GitHub stars
0Forks
2026-08-25Last updated
TypeScriptLanguage
MITLicense
Key features
- Instant upgrades, trend-gated downgrades — one smart verdict switches to the strong tier immediately
- A two-tier model router for DeepSeek Harness — automatic execution/judgment routing with an LLM Judge, multi-model fallback chains, exponential-backoff runtime failover, and task-level orchestration.
- Configuration changes — editing this profile patch (or the home patch) re-runs the affected plugin's apply() with the new config, no restart.
- Workers spawned by the subagent tool carry session.header.origin === 'subagent' and keep their pinned model
- coming back down requires a sliding-window majority (default 5 turns, ≥60%, low-confidence votes ignored).
Requirements
- The current Harness Web API proxy (@deepseek-ai/dsh-host-apiproxy) whitelists which settings namespaces the browser may read and write (WEB_SETTINGS_NAMESPACES).
- The official cards (shell, agent-loop, web-search-deepseek) are on the list
Install command
dsh plugin --profile web add github:green-dalii/dsh-shift-router