capture-learnings

Capture Learnings

After completing a task, capture non-obvious learnings that should persist to future conversations.

Process

1. Reflection Questions

Ask the user:

Surprises & Non-obvious Discoveries:

• "What was surprising or non-obvious about this task?"
• "Did you expect something to work differently?"

Mistakes & Corrections:

• "Did I make any mistakes that needed correction?"
• "Would you approach this differently next time?"

Validated Patterns:

• "Did any approach work particularly well?"
• "Is there a pattern here worth repeating?"

2. Categorize the Learning

If it's about...	Type	Example
How you prefer to work	feedback	"Don't reformat unrelated code"
A non-obvious behavior or gotcha	feedback	"Scout fixtures must be imported in specific order"
Current work context or deadline	project	"Feature freeze on 2026-04-01"
Where to find information	reference	"API metrics in Grafana at ..."

3. Save the Memory

Write to ~/.claude/projects/<project>/memory/knowledge/ with frontmatter:

---
name: feedback_[topic]
description: One-line description of the lesson
type: feedback
---

[State the rule/pattern clearly]

**Why:** [Reason — incident, constraint, or validated approach]

**How to apply:** [When/where this guidance kicks in]

Update MEMORY.md index with a new entry.

4. Escalation Check

If this is the second time correcting the same mistake:

• Escalate to rule in ~/.agents/rules/

If this is the third time:

• Escalate to skill in treadmill plugins

5. What NOT to Save

• Code patterns (derivable from reading code)
• Git history (use git log)
• Debugging solutions (the fix is in the code)
• Anything already in CLAUDE.md
• Ephemeral task details