description: "Route every decision through pre-trained GREEN/YELLOW/RED tiers and keep an audit trail — stage 2 of the agentic workflow." kind: "package-bundle"

@javierfrancoleal/decision-router

⚡ Super Agent Ready — part of the J.Franco governance layer for autonomous agents.

Created by J.Franco.

Summary

Stage 2 of the J.Franco agentic workflow:

Input → Detection (AI + human) → AI processes, decides, and proposes → Human reflects, verifies, and decides → Output

This plugin registers the decision-router skill on the host skill registry. The skill teaches the agent to assign one of three pre-trained tiers to every decision before acting:

• GREEN — pre-authorized: execute and log.
• YELLOW — propose the plan first and wait for approval.
• RED — escalate to the human; never act.

Training the router

The human trains the router by editing a policy file — decisions.md in the workspace root, .dsh/decisions.md, or ~/.dsh/decisions.md. Each row maps an action pattern to a tier. The file replaces the built-in defaults, so the human's training is always authoritative, and edits take effect on the next task.

Use this plugin

In Desktop, open Plugins and enable Decision Router. The skill appears in the session skill catalog; agents load it automatically when a TASK or WORKFLOW input needs decisions, and you can invoke it yourself by name.

How it works

• lib/index.js registers a bundled skill provider on ctx.skills, following the @deepseek-ai/dsh-skill-badge provider pattern.
• assets/decision-router.md holds the skill body.
• cordis.patch.yml inserts the provider row into the host plane of the Cordis tree, so its skills land in the global registry layer.

License

MIT — see the package license field.