Reflection Agent Pattern

Part 2 of 5 — critique-and-revise before delivery. Used in VAP Research + Architecture.

▶ Live demo · Architecture · Portfolio · VAP case study

What this is

Self-critique loops improve draft quality before ship — bounded reflection with trace viewer for teaching.

How we solve it

Critic agent revises generator output with stop conditions and structured traces — composable into VAP pipelines.

Case study & tradeoffs

venkat-ai.com/work · docs/ARCHITECTURE.md

---

Org skills: vpeetla-ai-skills. This repo includes .cursor/skills/, AGENTS.md, and CONTEXT.md.

git clone https://github.com/vpeetla-ai/vpeetla-ai-skills.git
./vpeetla-ai-skills/scripts/install.sh --cursor --codex --project .

---

Implementation status

Component	Status	Notes
Pattern demo + trace UI	✅	Live Vercel demo
Core agent loop	✅	Reference implementation
LangGraph production graph	🟡	Teaching scope — compose into VAP for fleet use
MCP tool bridge	❌	See LoopForge / VAP MCP docs
AegisAI gateway	❌	No side effects in pattern demo
Pytest regression	✅	pytest -q in repo

Part 2 of 5 in the Production Agent Patterns series.

Production-grade reference implementation of the Reflection pattern for accuracy-critical generation, code review, content validation, and complex reasoning workflows.

#	Pattern	Repository	Use when
1	ReAct	react-agent-pattern	Tool use + reasoning loops
2	Reflection	this repo	Self-critique and improve output
3	Plan-Execute	plan-execute-agent-pattern	Decompose goals into steps
4	Multi-Agent	multi-agent-system-pattern	Specialized role delegation
5	Swarm	swarm-agent-pattern	Parallel autonomous agents

▶ Live demo · 📖 Full series roadmap · 🚀 See in production — AI Content Factory

---

What you'll learn

• Draft → critique → improve → finalize loop
• Quality gates with score thresholds and retry limits
• Independent generator and critic interfaces
• Structured review findings for logging, UI, and evaluation

What this repo demonstrates

• Draft, critique, improve, and finalize loop
• Quality gates with score thresholds and retry limits
• Independent generator and critic interfaces
• Structured review findings that can be logged, displayed, or used for evaluation

Quick start

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
python -m reflection_agent_pattern
pytest

Runs without API keys using deterministic generator and critic stubs.

cp .env.example .env

See docs/LOCAL_DEVELOPMENT.md for production setup.

Repo layout

src/reflection_agent_pattern/
  __main__.py
  reflection.py
  models.py
  policy.py
docs/ARCHITECTURE.md
tests/test_reflection.py

Related

• Previous: ReAct Agent Pattern
• Next: Plan-Execute Agent Pattern
• Full pipeline: AI Content Factory

⭐ Star the repo if this pattern helps your work.