loop-engineering

A development pattern for agents that repeatedly observe results, choose actions, run tools, evaluate outcomes, and improve their settings. RAG means retrieving relevant information before generating an answer, while MCP is a way to connect agents to tools.

In plain words
What is it for?
Use it to build self-improving agents, adjust retrieval settings, connect approved tools, and store procedural memory for later runs.
Why use it?
It provides a defined way to improve agent performance using test scores, retrieved context, and lessons from earlier failures.

Skill for Claude CodeCodexCursor

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/vpeetla-ai/react-agent-pattern/loop-engineering
Any agent
npx skills add vpeetla-ai/react-agent-pattern --skill loop-engineering
Clone the repo
git clone --depth 1 https://github.com/vpeetla-ai/react-agent-pattern

Made for: Claude Code, Codex, Cursor.

Per session 48 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 307 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin 100% copy Near-identical to another mod in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00048 $0.00307
Opus 5 $0.00024 $0.00153
Sonnet 5 $0.00010 $0.00061
Haiku 4.5 $0.00005 $0.00031

Measured 2d ago against content hash 7f855b529f66, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

loop-engineering scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

Origin

This is a copy

100% identical to loop-engineering — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.

.cursor/skills/loop-engineering/SKILL.md · 43 lines

What it actually says

Loop Engineering

ODAEU outer loop

Phase Action
Observe Load RAG config vN, memory hints, corpus context
Decide Plan ReAct steps or orchestrator decomposition
Act MCP tools (read_file, search_docs) or repo patches
Evaluate Recall, faithfulness, pytest, review score
Update Tune RAG (top_k, hybrid_alpha) + write lesson

RAG evolve signals

Failure Tune
low_recall ↑ top_k, ↑ hybrid_alpha
low_faithfulness ↑ rerank_threshold, ↓ hybrid_alpha

Memory

  • Lesson(failure_mode, lesson, rag_version) → JSON store (v1)
  • Hints injected on next run via hints_for_query

MCP bridge

  • Local adapters in mcp/bridge.py; extensible to stdio MCP servers
  • Never give agents raw shell without allowlist

Reference

  • loop-engine-agent-platform/src/loop_engine/harness/
  • loops/support-intelligence.yaml
  • ADR-001 in LoopForge repo
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 43 lines · 48 tokens per session scan A 7f855b529f66

Subscribe to this mod's changes

loop-engineering is a skill published in the GitHub repository vpeetla-ai/react-agent-pattern (2 stars, last pushed 1mo ago), licensed MIT. It adds 48 tokens to every session and 307 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. It is 100% identical to loop-engineering, differing in 0 lines, and is treated as a copy.

Related

Other skills, from other repositories

deploy-vercel-render

Deploy vpeetla-ai demos: Vercel static/Next.js frontends, Render FastAPI backends, env vars, free tier gotchas. Use when shipping demos, fixing deploy failures, or adding render.yaml / vercel.json.

vpeetla-ai/multi-agent-system-pattern · 55 tokens

langgraph-orchestration

Build or modify LangGraph StateGraph agents in vpeetla-ai repos: typed state, nodes, conditional edges, MemorySaver, interruptbefore HITL. Use when adding orchestrators, coding loops, or multi-agent graphs.

vpeetla-ai/multi-agent-system-pattern · 52 tokens

portfolio-adr

Write ADRs, case studies, and portfolio copy for ai-architecture-portfolio and venkat-ai-portfolio. Use when documenting decisions, updating ecosystem pages, or syncing GitHub profile README with live demos.

vpeetla-ai/multi-agent-system-pattern · 46 tokens

tdd-agent-loops

Test-driven development for agent systems: red-green-refactor on graphs, mocked LLM fixtures, pytest-asyncio, trace assertions. Use when adding agent nodes, fixing loop bugs, or building pattern repos.

vpeetla-ai/multi-agent-system-pattern · 48 tokens

aegis-gateway

Integrate AegisAI gateway before tool side effects (notify, publish, deploy). Use when adding Slack/Telegram/WhatsApp notify, content publish, or any irreversible external action in VAP, AegisLoop, or ai-content-factory.

vpeetla-ai/multi-agent-system-pattern · 58 tokens

hitl-side-effects

Add human-in-the-loop gates for irreversible actions: LangGraph interruptbefore, AegisAI approval queue, UI approve/resume endpoints. Use when shipping, publishing, notifying, or merging agent-generated changes.

vpeetla-ai/multi-agent-system-pattern · 47 tokens