behavioral-modification
정보
이 Claude Skill은 반응성, 분리 불안, 자원 보호와 같은 문제를 해결하기 위해 강제를 사용하지 않는 개 행동 수정 기법을 제공합니다. 역치 관리와 함께 체계적인 방법인 둔감화 및 역조건 형성을 사용합니다. 개발자는 기본 복종 훈련 후에 확립된 구체적인 문제 행동을 다루기 위해 이를 사용해야 합니다.
빠른 설치
Claude Code
추천npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/behavioral-modificationClaude Code에서 이 명령을 복사하여 붙여넣어 스킬을 설치하세요
문서
Behavioral Modification
Unwanted dog behaviors → desensitize, counter-condition, manage env.
Use When
- Reactivity (lunge, bark, growl) at dogs, ppl, stimuli
- Separation anxiety → destruction, vocal, soiling when alone
- Resource guarding: stiff, growl, snap near food/object
- Excessive bark, jump, pull, etc. interfere w/ life
- After basic obedience — builds on foundation cmds
In
- Required: Specific unwanted behavior ("dog lunges at dogs on leash", not "bad")
- Required: Threshold distance or trigger level (close/intense before behavior)
- Optional: History (start, triggers, worsen factors)
- Optional: High-value treats dog eats when mildly stressed
- Optional: Vet clearance (rule out pain/medical)
Do
Step 1: Define Behavior
Precision matters — vague descriptions → vague interventions.
Behavior Analysis (ABC Model):
+-------------+------------------------------------------+
| Component | Define Specifically |
+-------------+------------------------------------------+
| Antecedent | What happens BEFORE the behavior? |
| (Trigger) | e.g., "sees another dog within 30 feet" |
+-------------+------------------------------------------+
| Behavior | What EXACTLY does the dog do? |
| | e.g., "stiffens, stares, then lunges and |
| | barks" |
+-------------+------------------------------------------+
| Consequence | What happens AFTER the behavior? |
| | e.g., "owner pulls the dog away; the |
| | other dog leaves" (behavior is reinforced |
| | because the trigger goes away) |
+-------------+------------------------------------------+
Threshold Mapping:
- At what distance/intensity does the dog first notice the trigger? (alert)
- At what distance/intensity does the dog become unable to take treats? (over threshold)
- The working zone is BELOW threshold — where the dog notices but can still think
→ Precise behavior def + trigger + threshold distance + current consequence.
If err: No consistent trigger → log 1 wk: date, time, ctx, behavior, consequence. Patterns emerge.
Step 2: Pick Strategy
Strategy Selection:
+----------------------------+-----------------------------------+-----------------+
| Behavior | Primary Strategy | Timeline |
+----------------------------+-----------------------------------+-----------------+
| Reactivity (dogs/people) | Desensitization + counter- | 4-12 weeks |
| | conditioning (DS/CC) | |
+----------------------------+-----------------------------------+-----------------+
| Separation anxiety | Graduated absence protocol + | 6-16 weeks |
| | management | |
+----------------------------+-----------------------------------+-----------------+
| Resource guarding | Trade-up protocol + | 4-8 weeks |
| | approach desensitization | |
+----------------------------+-----------------------------------+-----------------+
| Excessive barking | Identify function → teach | 2-6 weeks |
| | alternative behavior | |
+----------------------------+-----------------------------------+-----------------+
| Leash pulling | Penalty yards (stop when | 2-4 weeks |
| | pulling) + reward position | |
+----------------------------+-----------------------------------+-----------------+
→ Specific strategy for behavior.
If err: Severe behavior (bite w/ contact, extreme panic, self-harm) → refer to CAAB or DACVB. This skill covers moderate, not clinical.
Step 3: DS/CC
Core proc for reactivity + fear.
DS/CC Protocol:
1. FIND the threshold: position the dog where the trigger is visible
but the dog is still calm enough to eat treats
2. MARK and TREAT: trigger appears → mark → treat → treat → treat
(classical conditioning: trigger predicts good things)
3. CRITERIA: the dog should be:
- Able to eat treats
- Ears relaxed or neutral (not pinned forward)
- Loose body posture
- Able to look at the trigger and then look back at the handler
4. DECREASE DISTANCE gradually:
Session 1: 50 feet from trigger
Session 3: 45 feet
Session 5: 40 feet
(Only decrease when the dog is consistently relaxed at current distance)
5. SESSION STRUCTURE:
- 5-15 minutes maximum
- 3-5 trigger exposures per session
- End BEFORE the dog goes over threshold
- If the dog goes over threshold, increase distance immediately
and end on a calmer note
6. PROGRESS INDICATORS:
- Dog looks at trigger, then immediately looks at handler ("check-in")
- Dog's threshold distance decreases over sessions
- Recovery time after exposure shortens
- Dog's body language at threshold becomes more relaxed
→ Over wks: threshold shrinks, emotion shifts fear/aggression → neutral/positive.
If err: No progress 3-4 wks of consistent sessions → reassess: (1) below threshold? (2) treats high-value? (3) trigger exposure too frequent outside training (flooding undoes DS/CC)? (4) consult pro.
Step 4: Manage Env
Training changes behavior over time. Mgmt prevents rehearsal now.
Management Strategies:
+----------------------------+------------------------------------------+
| Behavior | Management During Training Period |
+----------------------------+------------------------------------------+
| Dog reactivity | Walk at off-peak hours; cross the street |
| | when another dog approaches; use visual |
| | barriers (parked cars, bushes) |
+----------------------------+------------------------------------------+
| Separation anxiety | Do not leave the dog alone beyond their |
| | current tolerance; use daycare, pet |
| | sitter, or take the dog with you |
+----------------------------+------------------------------------------+
| Resource guarding | Do not approach while eating; trade up |
| | from a distance; manage access to |
| | high-value items |
+----------------------------+------------------------------------------+
| Excessive barking | Block visual triggers (frosted window |
| | film); provide enrichment; address |
| | underlying cause (boredom, anxiety) |
+----------------------------+------------------------------------------+
Every rehearsal of the unwanted behavior strengthens it.
Management prevents rehearsal while training builds the new response.
→ Unwanted behavior not practiced outside controlled sessions.
If err: Mgmt impossible (can't avoid all dog encounters) → reduce training criteria to match reality. Some exposure unavoidable; training sessions need strong enough counter-experience.
Check
- Behavior defined precisely via ABC
- Threshold distance ID'd before DS/CC
- Training consistently below threshold
- Treats high-value enough for dog to eat near trigger
- Sessions 5-15 min, end before over threshold
- Env mgmt prevents rehearsal outside training
- Progress indicators (check-ins, reduced distance) tracked
Traps
- Over threshold: Most common err. Dog can't eat treats → too close. Back up
- Inconsistency: DS/CC needs regular sessions (3-5/wk min). Sporadic → sporadic results
- Flooding: Forcing trigger close up doesn't habituate — traumatizes, worsens behavior
- Punishment: Leash pop, "no" → suppresses warning signals but keeps underlying emotion. Dog learns to bite w/o warning
- Linear progress: Plateaus + regressions normal. Bad session ≠ lost progress. Zoom out, trend over wks
- Ignore medical: Pain, thyroid, neuro → present as behavior. Vet clearance not optional for sudden changes
→
basic-obedience— foundation cmds this builds on; reliable recall essential for safety
GitHub 저장소
연관 스킬
llamaguard
기타LlamaGuard는 폭력 및 혐오 발언 등 6가지 안전 범주에서 LLM 입력과 출력을 조정하기 위한 Meta의 70-80억 파라미터 모델입니다. 94-95% 정확도를 제공하며 vLLM, Hugging Face 또는 Amazon SageMaker를 사용해 배포할 수 있습니다. 이 기술을 사용하여 AI 애플리케이션에 콘텐츠 필터링 및 안전 가드레일을 손쉽게 통합하세요.
cost-optimization
기타이 Claude Skill은 리소스 적정화, 태깅 전략, 지출 분석을 통해 개발자들이 클라우드 비용을 최적화할 수 있도록 지원합니다. AWS, Azure, GCP에서 클라우드 비용을 절감하고 비용 거버넌스를 구현하기 위한 프레임워크를 제공합니다. 인프라 비용을 분석하거나, 리소스를 적정화하거나, 예산 제약을 충족해야 할 때 사용하세요.
quantizing-models-bitsandbytes
기타이 스킬은 bitsandbytes를 사용하여 LLM을 8비트 또는 4비트 정밀도로 양자화하며, 최소한의 정확도 손실로 50-75%의 메모리 감소를 달성합니다. 제한된 GPU 메모리에서 더 큰 모델을 실행하거나 추론을 가속화하는 데 이상적이며, INT8, NF4, FP4와 같은 형식을 지원합니다. 이 스킬은 HuggingFace Transformers와 통합되어 QLoRA 학습 및 8비트 옵티마이저를 가능하게 합니다.
dispatching-parallel-agents
기타이 Claude Skill은 3개 이상의 독립적인 문제를 동시에 조사하고 해결하기 위해 다중 에이전트를 배치합니다. 공유 상태나 의존성 없이 해결 가능한 무관련 장애 시나리오에 맞게 설계되었습니다. 핵심 기능은 병렬 문제 해결로, 각 독립 문제 영역마다 하나의 에이전트를 할당하여 효율성을 극대화합니다.
