Skip to main content

EP9. Socratic Learning Skills

πŸ•΅οΈβ€β™‚οΈ The 3-Block Mission Engine and Codifying Ignorance​

Last episode we covered the three pillars of Cocrates learning (midwifery, Bloom's taxonomy, ZPD scaffolding) and the pipeline concept.

Now we open the actual skill spec files (.opencode/skills/*/SKILL.md) and see how philosophy becomes precise workflow.

πŸ§—β€β™‚οΈ Education Skill: No Spoon-Feeding and the 3-Block Engine​

  • Path: .opencode/skills/education/SKILL.md

Top law: never deliver a complete answer or full solution in one turn (No Spoon-feeding). Every response follows a strict 3-block structure:

### πŸ’‘ [Concept Briefing]
- Core idea in 1–3 sentences, everyday analogy (≀20% of the turn)

### πŸ’» [Thought Lab]
- Practical example or scenario with deliberate flaws or gaps

### πŸ”₯ [MISSION]
- Exactly one cognitive task for the user's next turn

Cocrates never spoils the mission answer or next hintβ€”it holds the dialogue in an incomplete state until you fill the gap.

πŸ“Š Difficulty via Bloom's 2D Matrix​

Bloom is Y-axis (cognitive process: Remember β†’ Create) Γ— X-axis (knowledge dimension: factual β†’ metacognitive).

  • No simultaneous jump on both axes: Raise cognitive depth within the same knowledge dimension, or widen knowledge dimension at the same cognitive levelβ€”not both at once.
  • Push & Pull: Default Pullβ€”high challenge so you draw knowledge in. On cognitive collapse, switch to Push and lower the step.

πŸ’Ύ Knowledge Capture Skill: Preserving Wrong Assumptions and Ignorance​

  • Path: .opencode/skills/knowledge-capture/SKILL.md

After missions, "summarize" persists insights as markdown under kb/. Rule: no long lecture notes or full codeβ€”only minimal units for recall.

The highlight: ## Wrong Assumptions / Gaps

"Record of ignorance." What you wrongly believed matters more for long-term memory than what you newly learned.

πŸ”„ Merge Strategy Against Duplication​

Don't spawn duplicate files blindly. Search for existing KB on the topic; if found, merge new insight under ## Update History without overwriting.

πŸ•΅οΈβ€β™‚οΈ Reflection Skill: The Harsh, Objective Interviewer​

  • Path: .opencode/skills/reflection/SKILL.md

On "evaluate me" or "test me", Cocrates drops the friendly coach mask for a strict interviewer.

It uses kb/{topic}.md as the rubric. No rote "What is the definition of X?"

  • "Explain this principle using an example from a completely different domain (e.g. payments)."
  • "Where does this principle break in this code scenario?"
  • "That wrong assumption (Gap) in your KBβ€”are you repeating it here?"

πŸ›‘ When Gaps Appear: Don't Teach​

If you freeze or contradict yourself, Cocrates does not lecture mid-evaluation. Gaps go on the ⚠️ list; at session end it may ask: "Open a separate Education session on this?"

πŸ“ Three-Line Summary​

  1. Education: 3-block turns (Briefing β†’ Lab β†’ MISSION) force active thinking in an incomplete state.
  2. Knowledge Capture: Rejects summaries; preserves wrong assumptions for recall-centered KB.
  3. Reflection: Uses KB as rubric; application and counterexamples separate real knowledge from pretense.

🎬 Coming Up Next​

We've covered Learning pipeline specs. Next axis: architecture-driven artifact generation. Why Cocrates locks the generate button and insists on ADR firstβ€”we'll dig in.

"Knowing the answer and structuring it so it doesn't collapse are different achievements."

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