vstack¶

Organizational behavior, practiced on AI agents.

vstack is a curated library of 34 diagnostic patterns for AI agents and multi-agent systems, anchored in named organizational-behavior literature — Wharton's After-Action Review, Lencioni's Five Dysfunctions, Edmondson's Psychological Safety, Frei & Morriss's Trust Triangle, Schein's culture iceberg, Galbraith-Mintzberg structural fit — translated into runnable Python with public benchmarks and Substack-ready essays.

Most agent-observability tools capture what happened (traces). Most incident-response tools handle single events (a postmortem per alert). vstack ships a curated library of organizational practices — the same frameworks human teams use to learn, debate, escalate, and improve — implemented as patterns AI agents can run themselves.

Where the existing agent ecosystem treats failures as bugs to debug, vstack treats them as learning events to organize around.

Twelve ways to use it¶

Python imports — from vstack.lewin import LewinAttributionDetector
34 per-pattern CLIs — vstack-lewin analyze --trace trace.json
MCP server — vstack-mcp serve for Claude Desktop, Cursor, Cline, Continue, Zed, etc.
REST API — vstack-api serve (FastAPI on 127.0.0.1:8000) with auth, rate limiting, and observability baked in
Docker — docker run ghcr.io/valani9/vstack:0.7.0
Claude Code skills — 7 task-shaped SKILL.md files: /vstack-aar, /vstack-audit-crew, /vstack-post-incident, and more
Framework adapters — LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex, Pydantic AI
OpenAI Assistants + Anthropic Messages tool JSON — pure JSON, no install on the consumer side
Open WebUI plugin — drop-in tool manifest pointing at a running vstack-api
Tier B platform generators — Aider, Goose, Kiro, OpenClaw, Codex CLI, OpenCode (vstack-config gen-platform)
Browser dev tooling — vstack-browser scrapes agent traces from LangSmith / Phoenix / Helicone / Langfuse / Arize
First-run smoke (vstack-hello) — 30-second end-to-end demo with graceful no-key fallback

Three things to read next¶

Quick start — install + first detection in 60 seconds.
The 5-layer pattern shape — how every pattern is documented + implemented + demoed + benchmarked + written up.
Composition runbook — how the 34 patterns chain into real diagnostic workflows.

Quick install¶

pip install valanistack                # core library + 34 CLIs
pip install 'valanistack[anthropic]'   # + Anthropic client
pip install 'valanistack[mcp]'         # + MCP server
pip install 'valanistack[api]'         # + FastAPI REST surface
pip install 'valanistack[adapters]'    # + LangChain/LangGraph/CrewAI/LlamaIndex/Pydantic AI
pip install 'valanistack[all]'         # everything

Python 3.11+ required. MIT-licensed. Sole author: Ilhan Valani (valani@bu.edu).