kittors

agentflow

"AgentFlow — Multi-CLI agent workflow system with routing, safety detection, sub-agent orchestration, and knowledge graph memory."

kittors 14 1 Updated 3mo ago
GitHub

Install

npx skillscat add kittors/agentflow

Install via the SkillsCat registry.

SKILL.md

AgentFlow Skill

AgentFlow is a multi-CLI agent workflow system that keeps going until tasks are implemented and verified.

Features

  • 5-Level Routing (R0–R4): Every input is scored on 5 dimensions and routed proportionally — simple queries stay fast, complex tasks get full process, architecture-level tasks get dedicated review.
  • RLM Sub-Agent Orchestration: 6 specialized roles (reviewer, synthesizer, kb_keeper, pkg_keeper, writer, architect) plus native CLI sub-agents, dispatched based on task complexity.
  • EHRB Safety Detection: Three-layer safety (keyword scan, semantic analysis, tool output inspection) catches destructive operations before execution.
  • Three-Layer Memory: L0 user memory, L1 project knowledge base, L2 session summaries — context survives across sessions.
  • Knowledge Graph Memory: Graph-based project memory with progressive disclosure queries (AgentFlow unique).
  • Convention Extraction: Automatic coding pattern discovery from your codebase (AgentFlow unique).
  • Architecture Scanning: Proactive issue detection — large files, circular deps, missing tests (AgentFlow unique).

Quick Commands

Command Description
~init Initialize project knowledge base
~auto Auto-execute with full workflow
~plan Plan only, stop before development
~exec Execute existing plan
~status Show workflow status
~review Code review
~scan Architecture scan (AgentFlow unique)
~conventions Extract/check coding conventions (AgentFlow unique)
~graph Knowledge graph operations (AgentFlow unique)
~dashboard Generate project dashboard (AgentFlow unique)
~memory Manage memory layers
~rlm Sub-agent management
~validatekb Validate knowledge base consistency