intent-router-hardening

Rebalance recall and escalation for bilingual intent routing misclassifications.

Updated Aug 23, 2026
One-click install
npx skills add https://github.com/liuyu520/cc_source --skill intent-router-hardening
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: intent-router-hardening
Source: https://github.com/liuyu520/cc_source/tree/main/.claude/skills/intent-router-hardening
Command: npx skills add https://github.com/liuyu520/cc_source --skill intent-router-hardening

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Reusing existing intentRouter, localSearch, and prefetch hooks to tune simple-task detection and skill recall de-weighting in both Chinese and English, without hard-suppressing recall for mid-length requests.

Core Features & Use Cases

  • Reuses existing routing and recall controls to tighten boundaries between recall, execution, and model escalation.
  • Supports bilingual contexts (Chinese and English) with specialized gates and lookahead to avoid misclassification.
  • Facilitates safe, local-first hardening without introducing new global classifiers.

Quick Start

Adjust suppression weights in the intentRouter and localSearch layers to rebalance recall vs escalation for bilingual direct requests.

Frequently Asked Questions about intent-router-hardening

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fix bilingual intent routing misclassification for direct requests?

Balance recall and escalation in bilingual intent routing by de-weighting recall rather than hard-suppressing it for mid-length requests. Adjust suppression weights in intentRouter and localSearch layers to prevent misrouting direct queries to model escalation.

How do I de-weight recall suppression for mid-length requests in local search?

Use the SIMPLE_DIRECT_TASK gate with Chinese lookahead to enforce boundaries for simple tasks in bilingual contexts. Reuse existing intentRouter rules to ensure recall is de-weighted safely without misclassifying direct Chinese or English requests.

Does intent routing require a new global classifier for bilingual task classification?

Local-first hardening allows you to adjust IntentClass and TaskMode rules within the existing intentRouter and localSearch layers. This approach reuses current prefetch hooks to tighten boundaries without introducing new global classifiers.

What is the best way to adjust IntentClass and TaskMode rules for bilingual contexts?

Adjust IntentClass and TaskMode rules by reusing existing intentRouter and localSearch controls to tighten boundaries between recall, execution, and model escalation. Apply the SIMPLE_DIRECT_TASK gate with Chinese lookahead to avoid misclassification in bilingual contexts.

Why does hard-suppressing recall cause misrouting in bilingual intent routing?

Hard-suppressing recall causes misrouting in bilingual intent routing because it removes viable execution paths for mid-length requests. De-weighting recall instead of hard-suppressing it maintains the balance between recall, execution, and model escalation for bilingual direct requests.