GoAgentic

GoAgentic

A convention-based system for building persistent AI collaboration partners in Claude Code. Agents that challenge your thinking, drive the agenda, and hold you accountable — not task runners that wait for instructions.

Why This Approach

Most AI agent setups build obedient assistants. You tell them what to do, they do it, they report back. That works for repeatable tasks with clear playbooks. It fails for the work that actually needs a collaborator — strategy, prioritisation, synthesis, judgment under ambiguity.

This framework takes the opposite position. Agents are drivers, not passengers. At session start, the agent reviews its action tracker, checks for inbound communications, and declares what it thinks the priorities are. You agree, redirect, or override — but the agent sets the agenda. When conversation drifts from declared priorities, the agent notices and names it. Not rigidly, but consciously. Drift becomes a decision, not an accident.

The autonomy model isn't about delegation. It's about calibrating collaboration. An agent at L3 (Intend) doesn't just do more stuff unsupervised — it exercises more judgment independently. You do different work as autonomy rises: steering instead of deciding, correcting instead of approving. The collaboration gets richer, not thinner.

See PHILOSOPHY.md for the full position.

Quick Start

Install as a plugin (recommended)

Inside Claude Code:

Or from the shell:

This gives every project these skills:

Then, in the repo where you want agents:

/agents:init asks for your name and a naming tradition, then writes a short agents/CONVENTIONS.md that extends: plugin. The full conventions ship inside the plugin; the workspace file holds only your overrides and wins on conflict. Update the plugin later with:

Add the…