le-loop-stack 0.4.0

Everything for loop engineering — forge, score, run, benchmark, agent bridges

le-loop-stack

The only install you need for loop engineering — declare, validate, score, run, benchmark, and map any popular AI agent.

pip install "le-loop-stack>=0.4.0"

Includes:

Package	CLI	Role
le-loopforge	loopforge	Intent → LSS YAML, combine, LoopChain, loop library, export stubs
le-loopctl	loopctl / loop	Validate, score, pipeline, combine, quick, mix, bench
loopgym	(API)	Sim/Live/Replay + Loop Trace 1.0
loopbench	loopbench	Public leaderboard (19 tasks, 4 suites)

10-second start (token-efficient)

# Combine library loops → one compact YAML + JSON metadata
loop combine --library research-agent,autonomous-debugger \
  --intent "Research then fix tests" -o pipeline.yaml --json

# Or recipe mix (flatten + compact by default)
loop mix dev-agent --intent "Fix failing tests from CI" --json

# Single agent quick path
loop quick "Fix failing tests from CI" --agent aider

Optional extras: pip install "le-loop-stack[math,agents,bench]" — proof-carrying compose when loopmath installed.

Supported agents

loopctl agent list — langgraph, crewai, react, reflexion, dspy, aider, openhands, claude_code, codex, smolagents, autogpt, openai_agents

Optional runtime extras

pip install "le-loop-stack[agents]"   # LangGraph + CrewAI runtimes for export stubs
pip install "le-loop-stack[bench]"    # loopbench CLI
pip install "le-loop-stack[all]"

Full pipeline

loopctl pipeline --recipe dev-agent --intent "Repair CI" --suite suite-repair --compact --json

Golden Path: contributions/GOLDEN_PATH.md