An agent harness for orchestrating, managing, and extending AI agents across 18 repos. Not a code project --- a systems engineering workspace for agent workflows, cross-repo orchestration, and capability propagation.

Agentic Workflows

Quick Start ·  Workflow Tree ·  Features ·  Orientation ·  Ecosystem

Works with 

Quick Start

git clone https://github.com/B67687/agentic-workflows.git
cd agentic-workflows

Open AGENTS.md --- that's the operating contract (190 lines, 11 sections). The agent reads it first, then routes through the workflow tree.

bash scripts/test-smoke.sh                # 113-test smoke suite
bash scripts/infra/propagate-to-all.sh --apply  # Push templates to 18 sibling repos

Workflow Tree

The agent runs on a simplified single-loop pattern --- classify, execute, verify. The previous workflow engine (workflow.d/ YAML state machines) was stripped in consolidation; no workflow was ever active.

User request
  -> agentic loop (classify -> execute -> verify)
  -> tasks dispatch to sub-agents for context isolation

The workflow engine (workflow.d/ YAML state machines) was stripped in consolidation. No workflow was ever active (workflow-state.json always shows workflow: null). The agent now uses a simplified single-loop pattern: classify -> execute -> verify.

Features

🧠 Agentic Loop Single-loop classify -> execute -> verify. Sub-agents for context isolation. No YAML state machines.	📚 Skill System Debug, review, ship, document --- 44 engineering skills with progressive loading via MCP resources	🔄 Knowledge Propagation Change once in the hub, sync templates to 18 repos automatically via propagation contract
💾 Session State Single workflow-state.json for session tracking. No YAML state machine overhead.	⚡ Deterministic + Deliberative Split Scripts for deterministic operations (search, read, verify). AI + human for deliberative decisions.	🔬 Research Engine Frame -> discover -> triangulate -> apply -> preserve --- 6-phase methodology
🛡️ Human-Inspired Architecture Patterns from HumanLayer's daemon/session model --- session-scoped runtime, tree-based workflow state	🌐 Knowledge Propagation Sync config templates across repos via propagation contract	🧪 113-Test Smoke Suite Every change verified before commit. Covers tools, hooks, gates, pipelines, MCP, benchmarks.
🔁 Agentic Cycle Agent loops autonomously through tasks, using sub-agents for context isolation when needed.	📊 Session Observability Dashboard aggregates session state, archive history, and current context live	🧰 Organized Scripts Tools tracked in tools.toml. Clean separation: tools/, infra/, hooks/, research/.

Orientation

Common Commands

bash scripts/test-smoke.sh                 # 113-test smoke suite
bash scripts/tools/session-status.sh        # Workspace health
bash scripts/tools.sh                       # Tool registry (177 tools)
bash scripts/tools/session-dashboard.sh     # Live observability dashboard
bash scripts/tools/context-build.sh         # Context caching with three-layer strategy
bash scripts/tools/durable-run.sh           # Durable execution with checkpoint/resume
bash scripts/tools/handover.sh              # Session handover and state persistence
bash scripts/infra/propagate-to-all.sh --preview  # Preview template sync to 18 repos

See docs/workflow.md for the legacy reference. For full source-to-architecture traceability, see research/research-to-architecture-provenance.md.

Ecosystem

This harness was built by studying and integrating patterns from 80+ open-source projects across the agent ecosystem. Projects with * have patterns extracted into skills, scripts, or documentation.

Core Inspirations

AutoGen · *Google ADK · Claude Agent SDK · *OpenAI Agents SDK · *Pydantic AI · AutoGPT · MetaGPT · *A2A Protocol · *Hermes Agent · AgentScope · Open-SWE · *crewAI

*Claude Code · *Aider · *HumanLayer · *GStack · *Claude Plugins · UI-TARS · Deer Flow · *browser-use · *OpenCode · Pi · *SWE-agent · *DeepSeek-TUI · *9Router

*Agent-Skills · *12-Factor Agents · *Spec Kit · System Design Primer · *tree-sitter · *promptfoo

60+ projects across 9 more categories

Category	Projects
	*Mem0, LMCache, MemPalace, MemOS, PageIndex, *agentmemory, *GraphRAG, RAG-Anything
	*n8n, Flowise, Langflow, Dify, Manifest, Infisical
	Pi-Skills, *Karpathy-Skills, Codex Skills, *Counsel, Everything Claude Code, Awesome Claude Code, awesome-codex-skills
	MCP Registry, MCP Servers, GitHub MCP Server, sequential-thinking (structured reasoning), @opencode-ai/plugin
	Cline, CUA, *Rufo (ruflo), Agency-Agents, Codex CLI, generative-ai-for-beginners
	readme-svg-wave-divider-generator, readme-hub, GitHub Readme Stats, *awesome-design-md
	DeepSeek-V3, OpenAI Codex, Qwen, Gemini CLI, Hello Agents, Claude Code Best Practice
	Wiener (1948) Cybernetics · Meadows (2008) Thinking in Systems · Boyd (1976) Patterns of Conflict (OODA) · Grassé (1959) Stigmergy · Coram (2002) Boyd · O'Reilly (2023) Stigmergy & Future of Work

_{Ready to start? Read Quick Start or open AGENTS.md to begin.}

_· · ·

_{MIT License · Issues
Built with ♥ from the open-source agent community.}