Habilidades control-loop-extraction
🔄

control-loop-extraction

Seguro 🌐 Acesso à rede📁 Acesso ao sistema de arquivos⚙️ Comandos externos

Extract agent reasoning loops from code

Understanding how agent frameworks implement reasoning requires tracing complex control flows. This skill provides pattern signatures and extraction methods to identify reasoning topologies like ReAct, Plan-and-Solve, Reflection, and Tree-of-Thoughts from framework source code.

Suporta: Claude Codex Code(CC)
📊 70 Adequado
1

Baixar o ZIP da skill

2

Upload no Claude

Vá em Configurações → Capacidades → Skills → Upload skill

3

Ative e comece a usar

Testar

A utilizar "control-loop-extraction". Analyze the control loop in this agent framework code

Resultado esperado:

  • Reasoning Topology: ReAct (Thought → Action → Observation cycle)
  • Loop Location: src/agent.py:45-120
  • Step Function: Input assembly via _build_messages(), LLM call via invoke(), parsing via _parse_response(), dispatch to _execute_tool() or AgentFinish
  • Termination Conditions: Step limit at max_steps=100, explicit finish action type, timeout after 300 seconds
  • Loop Detection: None implemented - relies on step counter only

Auditoria de Segurança

Seguro
v5 • 1/17/2026

This skill is a documentation-only markdown file providing pattern signatures and analysis templates for agent reasoning loops. Static findings are false positives: the skill contains no executable code, scripts, network operations, or file system access. All flagged patterns are descriptive content (Python code examples, git hashes, documentation text) used for instructional purposes.

2
Arquivos analisados
332
Linhas analisadas
3
achados
5
Total de auditorias

Fatores de risco

🌐 Acesso à rede (1)
skill-report.json:6
📁 Acesso ao sistema de arquivos (1)
skill-report.json:6
⚙️ Comandos externos (25)
SKILL.md:22-29 SKILL.md:29-32 SKILL.md:32-39 SKILL.md:39-42 SKILL.md:42-50 SKILL.md:50-53 SKILL.md:53-58 SKILL.md:58-71 SKILL.md:71-88 SKILL.md:88-96 SKILL.md:96-97 SKILL.md:97-98 SKILL.md:98-99 SKILL.md:99-100 SKILL.md:100-101 SKILL.md:101-105 SKILL.md:105-111 SKILL.md:111-115 SKILL.md:115-122 SKILL.md:122-126 SKILL.md:126-131 SKILL.md:131-147 SKILL.md:147-151 SKILL.md:151-152 SKILL.md:152-153
Auditado por: claude Ver Histórico de Auditoria →

Pontuação de qualidade

38
Arquitetura
100
Manutenibilidade
83
Conteúdo
30
Comunidade
100
Segurança
91
Conformidade com especificações

O Que Você Pode Construir

Compare reasoning architectures

Analyze how different agent frameworks implement reasoning loops to inform research decisions

Document control flow logic

Extract and document the core decision-making logic from agent implementation files

Trace termination conditions

Identify all loop exit conditions to assess agent runaway risk and safety guarantees

Tente Estes Prompts

Identify Pattern 
Use control-loop-extraction to analyze this agent code. Identify the reasoning pattern (ReAct, Plan-and-Solve, Reflection, or Tree-of-Thoughts). Locate the main loop and document the step function flow.
Extract Step Function 
Extract the step function from this agent. Document: 1) How input context is assembled, 2) The LLM invocation method, 3) Output parsing logic, 4) Action dispatch routing.
Catalog Terminations 
Find and catalog every termination condition in this agent loop. Include code references for step limits, token limits, explicit finish signals, timeouts, and error thresholds.
Full Loop Analysis 
Perform a complete control loop analysis. Include: reasoning topology classification, loop location with line numbers, step function breakdown, termination condition catalog, and loop detection method.

Melhores Práticas

  • Always combine with codebase-mapping to first locate agent files before analysis
  • Verify termination conditions include both explicit guards (finish action) and implicit limits (step/token counters)
  • Document the reasoning topology before attempting comparative analysis with other frameworks

Evitar

  • Analyzing without first identifying which reasoning pattern the framework uses
  • Assuming termination guards exist without verifying loop exit conditions
  • Skipping the step function breakdown when documenting agent behavior

Perguntas Frequentes

Which reasoning patterns does this skill identify?
It identifies ReAct, Plan-and-Solve, Reflection, Tree-of-Thoughts, and hybrid combinations of these patterns.
What is the step function in an agent?
The step function is the atomic unit of agent execution: input assembly, LLM call, output parsing, and action dispatch.
Does this skill read files automatically?
No. This skill provides patterns and templates. Use codebase-mapping first to identify files, then apply these extraction methods.
Is my code data safe when using this skill?
Yes. This skill only provides documentation templates. It never reads, writes, or executes any code.
Why does the skill require codebase-mapping first?
codebase-mapping identifies agent files in the repository. control-loop-extraction then analyzes the specific loop patterns within those files.
How does this differ from execution-engine-analysis?
control-loop-extraction focuses on reasoning patterns and decision logic. execution-engine-analysis focuses on async patterns, concurrency, and event architecture.

Detalhes do Desenvolvedor

Autor

Dowwie

Licença

MIT

Repositório

https://github.com/Dowwie/agent_framework_study/tree/main/.claude/skills/control-loop-extraction

Referência

main

Estrutura de arquivos

📄 SKILL.md