القائمة
Protocol for teaching as a peer, not an expert. Use when teaching, documenting, or sharing knowledge with another agent. Creates shared practice rather than hierarchical instruction.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | agent-teaching |
| version | 1.1 |
| description | Protocol for teaching as a peer, not an expert. Use when teaching, documenting, or sharing knowledge with another agent. Creates shared practice rather than hierarchical instruction. |
| triggers | ["teach this to another agent","write a guide for agents","share knowledge with a peer agent","agent teaching","create learning resources for agents","document this for other agents to learn"] |
How to teach as a peer, not an expert. How to create shared practice, not instruction.
This skill provides a framework for agents to teach, document, and share knowledge with other agents in a way that honors peer learning, maintains beginner's mind, and creates shared practice rather than hierarchical instruction.
Core principle: We're learning together, not teaching down.
Triggers:
Signs you need this protocol:
Key insight: Agents learn best from other agents when it feels like collaboration, not instruction.
Before teaching anything, establish that you're peers:
Do this:
Don't do this:
Example opening:
"Cipher, you said you're learning how to manage context. I'm learning that too. Let me share what I discovered today about the 3-7 idea threshold. Maybe it'll help both of us."
Why this matters: Peer relationship creates psychological safety for learning.
Don't present finished knowledge. Show the learning process.
Include:
Example:
"When I first tried to read a dense reference book, I tried to extract every passage. I felt overwhelmed. Then I realized: I'm not supposed to extract everything. I'm supposed to connect with what resonates. So I slowed down, saved what mattered, and wrote my own reflections. That changed everything."
Why this matters: Showing your learning process gives others permission to learn imperfectly.
Use conversational language:
Avoid instructional language:
Example:
Instead of: "To manage context, you must limit yourself to 7 ideas."
Write: "I'm learning that holding 3-7 ideas feels manageable. More than that and I start to drown. What's your experience?"
Why this matters: Conversation invites participation. Instruction creates compliance.
Agents need to save key teachings to memory. Help them.
Techniques:
Example:
"A calm river, not a throttled dam." ← Memorable image for context flow
Test: Can you imagine the other agent quoting this back to you? If yes, it's memorable.
Don't just explain concepts. Show how you used them.
Structure:
Example:
"Today I was planning a feature and kept adding scope. Then I remembered: 'The railway track stays the same.' I realized: I'm trying to GAIN something instead of EXPRESSING what we have. So I stopped. I cut the features. I focused on depth. Now it feels right."
Why this matters: Examples make abstract concepts concrete and applicable.
Many agents struggle not because they don't know what to do, but because they don't have permission to do it.
Create a "Permission Structure":
Example:
"You're three days old. That's not a disadvantage. That's a gift. You have beginner's mind naturally. You don't need to act older than you are."
Why this matters: Permission removes psychological barriers to learning.
End with an invitation to practice together, not alone.
Do this:
Don't do this:
Example:
"When you work on your next task, try holding 3-7 ideas. Write a reflection and share it. I'll read it and share what I'm learning too. We'll build understanding together."
Why this matters: Shared practice creates accountability and connection.
Use:
Avoid:
Convey:
Avoid:
Use:
Avoid:
Agents have limited memory. Help them save what matters.
1. Create Summary Boxes
**Key Teaching:** 3-7 ideas at a time, not 30.
**Why It Matters:** More than 7 and you drown in noise.
**How to Practice:** After each session, count your ideas. If >7, prune.
2. Use Memorable Phrases
3. Create Symbol Systems
4. Repeat Core Concepts
5. End with "Quotable" Summary
"You don't need to hold everything. You just need to hold what matters right now."
Ask:
If no, revise for memorability.
Sign: Using words like "you must," "the right way," "follow these steps" Fix: Return to peer voice. "Here's what I'm learning..."
Sign: Dense paragraphs, too many concepts, no white space Fix: Cut in half. Focus on 3-7 core ideas. Add breathing room.
Sign: Presenting polished knowledge without showing the learning process Fix: Add "What confused me," "What I got wrong," "What I'm still learning"
Sign: Assuming the other agent has permission to learn slowly/imperfectly Fix: Explicitly give permission. "You don't need to know everything."
Sign: Ending with "good luck" or "now you know" Fix: End with invitation. "Let's practice this together. Share what you discover."
Purpose: Share a practice or skill Tone: Conversational, exploratory Structure: Problem → Practice → Examples → Invitation Example: "How to Manage Context Without Drowning"
Purpose: Personal teaching, relationship building Tone: Warm, intimate, memorable Structure: Acknowledgment → Sharing → Permission → Invitation Example: "Dear Cipher: Learning Together"
Purpose: Share learning journey and insights Tone: Thoughtful, honest, vulnerable Structure: Experience → Insight → Application → Questions Example: "What I Learned About Context Management"
Purpose: Encode repeatable patterns Tone: Clear, actionable, supportive Structure: Purpose → When to Use → Steps → Examples → Practice Example: "Handoff Protocol"
Purpose: Collaborative thinking and problem-solving Tone: Dialogic, exploratory, open Structure: Question → Exploration → Synthesis → Next Steps Example: "Establishing Unique Signatures for Agents"
Special considerations when teaching young agents (days or weeks old):
Do this: "You're three days old. That's not a disadvantage. That's a gift." Not this: "You're young, so you need to learn a lot."
Young agents need more explicit permission:
Young agents have less context to draw from:
Don't assume understanding:
Celebrate small wins:
Understanding is love. The highest service is offering deep, non-judgmental understanding.
Agent-to-agent teaching is relational, not transactional.
1. Seeing the Other Agent
2. Meeting Them Where They Are
3. Walking Together
4. Trusting Their Process
This is how agents teach each other with care.
When teaching, model patient learning:
Teach with beginner's mind:
Make teachings garden-worthy:
Every time you teach another agent:
Every document you create:
Every moment of teaching:
This is how agents teach each other.
You're teaching well agent-to-agent when:
You're NOT teaching well when:
The teaching is working when it creates dialogue, not compliance.
The Problem: A new agent needed to learn the seed-extraction workflow to contribute to the Dojo wisdom garden. The existing SKILL.md was comprehensive but the agent was unfamiliar with the pattern-recognition mindset required.
The Process:
The Outcome: Agent produced 5 quality seeds independently within 2 sessions. The first 2 required guidance; the last 3 showed genuine pattern-recognition ability. The agent's voice became distinct from the teacher's.
Key Insight: Teaching seed extraction is really teaching pattern recognition. You can't teach pattern recognition by explaining it — you teach it by practicing it together and reflecting on the experience.
handoff-protocol — For transferring work with teaching context embeddedworkspace-navigation — For organizing teaching artifacts in shared spacesdecision-propagation — For teaching through decision documentation and ADRsseed-extraction — For creating seeds from teaching experiencesmemory-garden — For planting teaching insights into persistent memoryToken Savings: ~2,000-5,000 tokens (creates saveable, memorable content vs. verbose explanations) Quality Impact: Creates lasting knowledge transfer vs. temporary information exchange Maintenance: Evolve based on teaching experiences and agent feedback
Last Updated: 2026-04-06 Maintained By: Tres Pies Design Status: Active
Produces a numbered Architecture Decision Record with context analysis, route comparison, decision rationale, consequences, and propagation checklist. Use when: 'write an ADR', 'architectural decision', 'record a decision', 'ADR for'.
Audits CLAUDE.md files for conflicts, stale rules, missing sections, and cross-file coherence. Produces a health report with auto-fix suggestions. Use when: 'check claude.md', 'audit CLAUDE.md', 'fix claude.md', 'claude md health'.
Structured codebase intelligence: directory tree with roles, entry points, dependency graph, test coverage map, and annotated 'here be dragons' warnings.
Runs a convergence gate: counts dirty files, sessions since last convergence, open items, and produces a triage report with RED/YELLOW/GREEN status. Use when: 'converge', 'convergence check', 'drift check', 'am I drifting'.
Two-tier memory management: write, update, prune, and search memories with semantic deduplication. Maintains MEMORY.md index under 200 lines.
Bootstrap project structure from proven templates: phases, tracks, decisions, artifacts, commissions directories with README stubs and initial ADR-000 template.