AI coding

Routing requests and skills to agents

• A Trigger Regression Set for Skills: Sampling and Pass Criteria
• Building a Tie-Breaker Agent: Rules and Risks
• Routing to Nothing: When the Right Answer Is No Skill
• When a Router Picks a Tool the Agent Cannot Call
• Writing a Skill Router: How Models Choose Between Skills
• A Canary Description for Skill Routing: Shadow Loads
• A Confidence Band That Survives a Model Swap
• A Confusion Table per Skill: Reading False Loads and Misses
• A Permission Check Between the Router and the Tool Call
• A Router Log Schema: Fields for Reconstructing a Skill Choice
• A Rule-Based First Pass: When a Deterministic Check Beats a Model
• A Skill Load Test: Measuring Trigger Accuracy Across Requests
• A Trigger Accuracy Metric the Router Logs on Every Skill Call
• A Trigger Regression Set: Sampling and Pass Criteria
• False Fires Cost Tokens: Pricing a Skill Trigger Error
• Filtering a Tool Catalog by Role Before the Prompt Is Built
• From Queue Resolution to Router Rule: A Change Log
• From Shadow Result to Cutover: A Switch Checklist
• Holding Requests Fixed When You Compare Routers
• How to Name a Skill So the Router Picks It
• How to Test That a Skill's Trigger Fires
• Reading a Router Log to Find Descriptions That Need Splitting
• Reading a Router's Second Choice When the First Is Blocked
• Routing Tool Calls Across Models and Providers
• Stratifying a Traffic Sample for a Model Comparison
• The None Option in a Router Prompt: Wording and Order
• Two Skills Overlap: When to Merge, Split, or Add a Disambiguator
• When a Router Sends a Request to Two Skills: Disambiguation
• Who May Edit the Router Prompt and the Skill Registry
• Who Owns a Router Miss: Routing It to a Skill Author