bianhaifeng789-hue/skills-experimental/classifier-denial-tracking-pattern

Classifier Denial Tracking Pattern Skill

Classifier Denial Tracking Pattern - DenialTrackingState consecutiveDenials/totalDenials + DENIAL_LIMITS maxConsecutive:3/maxTotal:20 + recordDenial increment + recordSuccess reset consecutive + shouldFallbackToPrompting limits check + consecutive≥3 OR total≥20 fallback + immutable state update + classifier retry budget。

功能概述

从Claude Code的utils/permissions/denialTracking.ts提取的Classifier denial tracking模式，用于OpenClaw的分类器拒绝计数。

核心机制

DenialTrackingState consecutiveDenials/totalDenials

export type DenialTrackingState = {
  consecutiveDenials: number
  totalDenials: number
}

export function createDenialTrackingState(): DenialTrackingState {
  return {
    consecutiveDenials: 0,
    totalDenials: 0,
  }
}
// DenialTrackingState
# consecutiveDenials: number
# totalDenials: number
# Immutable state

DENIAL_LIMITS maxConsecutive:3/maxTotal:20

export const DENIAL_LIMITS = {
  maxConsecutive: 3,
  maxTotal: 20,
} as const
// DENIAL_LIMITS
# maxConsecutive: 3
# maxTotal: 20
# as const

recordDenial increment

export function recordDenial(state: DenialTrackingState): DenialTrackingState {
  return {
    ...state,
    consecutiveDenials: state.consecutiveDenials + 1,
    totalDenials: state.totalDenials + 1,
  }
}
// recordDenial
# consecutive + 1
# total + 1
# Immutable update

recordSuccess reset consecutive

export function recordSuccess(state: DenialTrackingState): DenialTrackingState {
  if (state.consecutiveDenials === 0) return state // No change needed
  return {
    ...state,
    consecutiveDenials: 0,
  }
}
// recordSuccess
# consecutive = 0
# No change if already 0
# Reset on success

shouldFallbackToPrompting limits check

export function shouldFallbackToPrompting(state: DenialTrackingState): boolean {
  return (
    state.consecutiveDenials >= DENIAL_LIMITS.maxConsecutive ||
    state.totalDenials >= DENIAL_LIMITS.maxTotal
  )
}
// shouldFallbackToPrompting
# consecutive >= 3
# total >= 20
# Fallback to ask

consecutive≥3 OR total≥20 fallback

// After 3 consecutive denials or 20 total denials
// the classifier has failed too often → fallback to prompting
if (shouldFallbackToPrompting(state)) {
  // Show permission prompt instead of using classifier
}
// consecutive≥3 OR total≥20
# 3 consecutive = classifier failing repeatedly
# 20 total = classifier unreliable overall
# Fallback to prompt user

immutable state update

// All functions return new state objects
// Original state is never mutated
return { ...state, consecutiveDenials: state.consecutiveDenials + 1 }
// immutable state update
# Spread + update
# No mutation
# Functional update

classifier retry budget

// Budget pattern:
// - 3 consecutive: tight budget - classifier repeatedly failing on similar operations
// - 20 total: loose budget - classifier unreliable over session lifetime
// Budget exhausted → fallback to explicit user prompts
// classifier retry budget
# 3 consecutive = tight
# 20 total = loose
# Budget exhausted → prompt

实现建议

OpenClaw适配

1. denialState: DenialTrackingState type pattern
2. limits: DENIAL_LIMITS constants pattern
3. recordDenial: increment counters pattern
4. recordSuccess: reset consecutive pattern
5. fallback: shouldFallbackToPrompting pattern

状态文件示例

{
  "consecutiveDenials": 2,
  "totalDenials": 8,
  "maxConsecutive": 3,
  "maxTotal": 20,
  "shouldFallback": false
}

关键模式

Consecutive/Total Dual Counters

consecutiveDenials + totalDenials → dual counters → consecutive = tight budget | total = loose budget → dual counters
# consecutive/total dual counters
# consecutive: repeated failures
# total: overall reliability

3 Consecutive Tight Budget

maxConsecutive: 3 → 3 consecutive denials → classifier failing repeatedly → tight budget → fallback
# 3 consecutive tight budget
# repeated failures
# immediate fallback

20 Total Loose Budget

maxTotal: 20 → 20 total denials → classifier unreliable overall → loose budget → fallback
# 20 total loose budget
# session lifetime budget
# overall reliability

recordSuccess No-Change Optimization

if (state.consecutiveDenials === 0) return state → no change needed → ref equality preserved → optimization
# recordSuccess no-change optimization
# early return if 0
# preserve ref equality

Immutable Functional Update

{ ...state, consecutiveDenials: state.consecutiveDenials + 1 } → spread + update → immutable → functional → no mutation
# immutable functional update
# spread + increment
# return new object

倉用价值

• ⭐⭐⭐⭐⭐ Consecutive/total dual counters pattern
• ⭐⭐⭐⭐⭐ 3 consecutive tight budget pattern
• ⭐⭐⭐⭐⭐ 20 total loose budget pattern
• ⭐⭐⭐⭐ recordSuccess no-change optimization pattern
• ⭐⭐⭐⭐⭐ Immutable functional update pattern

来源

• Claude Code: utils/permissions/denialTracking.ts (43 lines)
• 分析报告: P59-8