kinhluan 🧠 Second Brain Reflection & Compression
"He who learns but does not think is lost! He who thinks but does not learn is in great danger." — Confucius (adapted for Agents)
This skill transforms raw experience into compressed, actionable tri thức (knowledge). It ensures that every failure is a lesson and every success is a reusable pattern.
🔄 The Reflection Protocol
Trigger this protocol after completing a Mission, PR, or Experiment.
1. The Art of War Audit (Ngũ Sự)
Evaluate the task through the lens of Sun Tzu's Five Factors:
- Tao (Alignment): Did our output match the original
why-strategic-rationale? - Heaven (Context): Did external shifts (market, new library updates) affect us?
- Earth (Terrain): What new technical debt or "architectural traps" did we find in the codebase?
- Command (Leadership): Were our decisions data-backed? Where did we guess?
- Method (Discipline): Which parts of our SDLC (tests, lint, deploy) felt slow or fragile?
2. Knowledge Extraction
Identify 1-3 critical insights. Ask:
- "What was the 'Aha!' moment?"
- "If I had to do this again from scratch, what would I do differently in the first 10 minutes?"
- "What is the one rule that would have prevented our biggest hurdle in this task?"
🗜️ The Compression Engine (Terse Format)
Do NOT save conversational fluff. Use the Rule/Stratagem Format:
- Rule:
Condition/Context → Action/Constraint. (Reason). - Stratagem:
Pattern → Optimization. (Impact).
Examples:
- Bad: "We realized that when we use the standard Python json library it is too slow for 1GB files, so we should use orjson next time to save 5 seconds."
- Good (Compressed):
Rule: Large JSON (>500MB) → Use orjson, not stdlib. (5x faster serialization).
💾 Memory Tiering (Where to Save)
Agent must determine the scope before saving:
| Scope | Location | Action |
|---|---|---|
| Project Specific | ./GEMINI.md |
Add to "Shared Rules" section. |
| Private/Task Finding | ~/.gemini/tmp/skills/memory/MEMORY.md |
Add to local project index. |
| Universal Truth | ~/.gemini/GEMINI.md |
Add to "Personal Preferences" section. |
🚦 Output Format
After processing, the agent should present the knowledge in this format:
### 🧠 Second Brain Sync — [Task Name]
**1. Strategic Audit (Sun Tzu):**
- [Factor]: [Finding]
**2. Compressed Knowledge:**
- `Rule: [Context] → [Action]. ([Why]).`
**3. Saving Location:**
- [Path to file]
🔄 Workflow Integration — When to Trigger
Second Brain Reflection is a closing ritual for every major workflow. Trigger automatically at the end of:
PhD/Research Workflow Closing Ritual
After completing each stage:
- Post-survey: What search strategies worked? Which venues had the best signal?
- Post-RQ formulation: What gaps were most fruitful? What RQ patterns to avoid?
- Post-experiment: Which hyperparameters mattered most? What debugging sequence saved time?
- Post-paper submission: What reviewer feedback was predictable? What writing pattern to repeat?
- Post-defense: What committee questions were surprising? What slides worked best?
MVP/Product Workflow Closing Ritual
After completing each stage:
- Post-problem-discovery: Which validation method gave the strongest signal? What interview question was most revealing?
- Post-WHY statement: What assumption was most wrong? What stakeholder pushback was most valuable?
- Post-design (C4/DDD): What boundary was hardest to draw? What Generic Subdomain should have been SaaS?
- Post-ship: What CI/CD step failed most? What test would have caught the production bug?
- Post-release: Which Rogers gate had the weakest signal? What feature flag strategy worked best?
Ngũ Sự Assessment Closing Ritual
After each strategic assessment:
- Which factor was most surprising? (usually Earth or Command)
- What score changed the most from last assessment?
- What action item from last assessment was actually completed?
Dev/Task Closing Ritual
After major coding tasks:
- What pattern should become a project rule?
- What tool/library saved the most time?
- What anti-pattern appeared that should be banned?
Reflection Gate
Before marking any task as DONE in collaborative-engineering-agent Kanban, ask:
"What is the one rule from this task that would save time on the next similar task?"
If the answer is non-trivial, run Second Brain Reflection before closing the task.
🔗 Integration
- →
art-of-war: Provides the Audit framework. - →
master-framework: Lessons feed back into Layer 0 (WHY) and Layer 2 (DESIGN). - →
collaborative-engineering-agent: Reflection Gate before marking tasks DONE. - →
kinhluan-router: Reflection results update routing heuristics ("last time X happened, Y skill was most useful").