Skip to main content

EP13. Using Cocrates Harness Well

🌐 The Feedback Loop That Evolves Together and Open Source Spirit​

You've installed Cocrates, run the exercises, and studied the architectural principles underneath.

Is learning over? No. The real journey starts now. Understanding architecture is like studying the engine—now you drive and build feel on the road.

This finale episode: how user and tool co-evolve, and what extensibility really means.

1. The User Must Evolve: Harnessing Ignorance​

The thread through this series is simple: break the habit of accepting AI output unexamined into a black box. Examination means understanding through questions, judging with agency, and approving explicitly.

  • Get comfortable with dialogue: Cocrates won't dump answers—it keeps asking. "Can't it just tell me?" is natural. Those questions are a partner handing agency back to you.
  • Internalize the questions: Over time, Socratic questions from outside become your own thinking habits. "What's the trade-off?" "What constraint am I missing?" When you ask yourself, you evolve into an irreplaceable planner and builder in the AI era.
  • Habituate review and approval: You scrutinize every line in a PR—why copy-paste AI output raw? AI deliverables are review targets too. Can you say yes to "Does this match the request?" and "Can I explain this architecture to someone else?"

2. Cocrates Harness Must Evolve Too​

Cocrates isn't a sealed factory product—it's an open blueprint you can rework anytime.

🛠️ Extensible Skills​

Bundled skills are a starting point. Add or extend skills for your repeating patterns:

  • Frequent weekly reports → weekly-report-authoring
  • Many proposals → proposal-authoring
  • Team onboarding, deploy checks—custom skills

Via generating-skill-creation, "create a skill that generates ~~" designs reusable skill architecture for you.

✍️ Custom Process to Taste​

SKILL.md files under ~/.config/opencode/skills/ are markdown—anyone can read and edit.

  • "Review steps are exhausting" → simplify stages
  • "Doesn't match how I work" → reshape workflow

🔄 A Closed Feedback Loop That Grows Together​

The more rigorously you review, the denser the agent's output; better output grows your judgment. Participation and performance scale together.

Experience → improve agent/skills → better experience → user growth

Like training wheels on a first bike, Cocrates paces early architecture—then, when you balance, becomes a companion offering deeper business challenges.

🌐 Open Source Philosophy That Lives Everywhere​

Cocrates source, prompts, and architecture docs are transparent.

Official repository: https://github.com/cocrates/cocrates.ai

Transparency means more than sharing code—platform independence. Socratic dialogue and Snowflake workflow aren't tied to one IDE or plugin.

Claude Code, Cursor, Codex, Antigravity—wherever agents run, you can transplant this blueprint and build your harness. Best evolution: absorb the spirit and craft your own custom harness—not only use the shipped tool.

🎯 Principle of Continuous Improvement​

One absolute constitution holds the architecture worldview:

"The unexamined code is not worth generating."

It applies not only to agent output but to Cocrates Harness itself—prompts, skills, workflows. Prefer steady small improvements over one-shot perfect skills.

🎬 Coming Up Next​

Thirteen episodes on taking agency and growing with AI agents architecturally. You understand principles, hold the tool, and lit the fuse of evolution.

One step remains.

Next isn't a manual—it's a milestone to carry in your chest: "Cocrates Harness User Manifesto — Never Stop Growing."

This series introduces the Cocrates Harness framework. Cocrates is an agent harness designed for Socratic dialogue so users keep agency and grow.