Post-Mortem & Skill Evolution

When to use this skill

• •After a complex debugging session (debugging-code completion).
• •At project completion (Project Retrospective).
• •When the user asks "Why did this keep happening?" or "Update your protocols".
• •When you detect a recurring pattern of errors related to a specific tool.

Philosophy: The Self-Evolving Agent

"A mistake repeated is a decision."

This skill transforms Reactive Fixes (fixing the code) into Proactive Guardrails (fixing the process).

Workflow

1. Root Cause Analysis (RCA)

Analyze the recent interaction history or debugging-code logs.

• •Incident: What broke? (e.g., "CUDA OOM error")
• •Root Cause: Why? (e.g., "Batch size 64 was too large for 8GB VRAM")
• •Pattern: Is this governed by a using-* skill? (e.g., using-ultralytics)

2. Gap Analysis

Check the current SKILL.md for the relevant tool.

• •Does it already warn about this?
• •Was the warning ignored, or was it missing?

3. Optimization Proposal

Draft an update to the SKILL.md to prevent this specific failure mode.

• •New Check: Add to "Pre-Flight Checks".
• •New Reference: Add to "Quick Reference".
• •New Rule: Add to instructions.

4. Implementation

1. •Draft: Create the specific markdown update.
2. •Verify: Ensure it doesn't conflict with existing rules.
3. •Apply: Use replace_file_content to update the SKILL.md.

Example Scenario

Incident: User faced RuntimeError: dataset.yaml not found when training YOLO. Analysis: The using-ultralytics skill mentions datasets but doesn't force a path check. Action:

1. •Open skills/using-ultralytics/SKILL.md.
2. •Add a check to "Pre-Flight Checks":

   "Verify data YAML path exists and is absolute before passing to model.train()."

3. •Notify User: "I've updated my internal protocols to always verify YAML paths before training in the future."

Trigger Words

• •"Conduct a post-mortem"
• •"Update your learnings"
• •"Never let this happen again"