Status: Proposed
Date: 2026-06-30

Context

The SkillLens research framework from Microsoft defines reusable data
structures for classifying agent behaviour patterns: Mode (success/failure
pattern with evidence) and ModeSet (collection of modes with source
trajectories). The skill-quality-auditor’s analysis pipeline currently lacks
a structured way to represent and pass around these patterns.

Decision

Port the Mode and ModeSet types from SkillLens into a new
analysis/modes.go file:

type ModeType string
const (
    SuccessMode ModeType = "success"
    FailureMode ModeType = "failure"
)

type Mode struct {
    Type                ModeType
    Pattern             string
    Description         string
    Evidence            string
    SourceTrajectoryIDs []string
}

type ModeSet struct {
    SuccessModes        []Mode
    FailureModes        []Mode
    SourceTrajectoryIDs []string
    Summary             string
}

Also extract SkillLens’s failure mode categories (error patterns,
anti-patterns, pitfalls) into the D3 anti-pattern scorer as additional
detection patterns.

Consequences

• Low-effort, high-value port with immediate integration points
• Duplication analysis pipeline can output ModeSet for richer reports
• D3 scorer gains empirically grounded categories from 5 benchmarks
• Meta-skill findings inform rubric documentation in framework-dimensions.md