notque/vexjoy-agent

VexJoy AI Agent with Intelligent Routing - /do routes plain-English requests to the right specialist agent and gates the work with reviews, tests, and a learning loop.

What it solves

VexJoy Agent prevents AI agents from "skipping steps" or shipping broken code by replacing advisory instructions with structural enforcement. It solves the problem of agents confidently claiming work is complete or "trivial" without actually running tests or verifying changes, ensuring that high-quality methodology is followed regardless of the agent's confidence level.

How it works

The toolkit operates through a four-layer architecture that integrates directly into AI CLI runtimes (such as Claude Code, Codex, Factory, and Reasonix):

  • Agents: Provide domain-specific knowledge, including failure mode catalogs and error-to-fix mappings.
  • Skills: Define phased methodologies with strict exit criteria. An agent cannot move to the next phase without providing evidence of success.
  • Hooks: Lifecycle event triggers that block incomplete work at zero LLM cost.
  • Scripts: Deterministic tools (linters, test runners) that handle the actual execution and verification, removing LLM judgment from the verification process.

Users interact with the system via a router (e.g., using the /do command), which analyzes the intent and automatically pairs the correct specialist agent with the appropriate skill and methodology.

Who it’s for

  • Developers: Who need an agentic workflow that guarantees code verification and prevents regressions.
  • Knowledge Workers: Who use the toolkit for research, drafting, and creating self-contained HTML artifacts (reports, slide decks, prototypes) without writing code.
  • AI Power Users: Who want to customize routing tables, pipelines, and hooks.

Highlights

  • Anti-Rationalization: Uses exit gates and hooks to block agents from skipping tests or verification, demanding exit codes rather than assertions.
  • Multi-Runtime Support: Compatible with Claude Code, Codex, Factory, and Reasonix.
  • Harness-Integrated Routing: Automatically routes plain-English requests to the right specialist agent and methodology.
  • Knowledge Work Engine: Includes a content engine that validates drafts against AI patterns and generates high-fidelity HTML artifacts.
  • Evidence-Based Evolution: New skills are A/B tested against baselines before being merged into the toolkit.