save-learning

---

name: save-learning description: Save a new learning, pitfall, or coding directive to the correct file with cross-file deduplication. Routes stack-specific pitfalls to the appropriate {stack}/CLAUDE.md "Known Pitfalls" section and cross-cutting patterns to memory/MEMORY.md Key Learnings. Use when you've learned something new from a correction, bug, or debugging session and want to persist it correctly. Trigger when: tech-writer agent needs to save a learning, /directive needs to write a new directive, or mid-session you discover a pitfall worth recording. Examples: "save this learning", "record this pitfall", "add this directive", "write this to the right CLAUDE.md".

save-learning

Single entry point for saving learnings, pitfalls, and directives to the correct file. Callers provide the pre-written text — this skill handles routing and deduplication only.

Separation of concerns:

• Caller (tech-writer, /directive, mid-session) = writing quality (crafts the text, code examples, phrasing)
• This skill = placement intelligence (classifies, deduplicates, writes to correct file)

---

Step 1: Classify the learning

If the caller provided an explicit stackHint (backend / web / mobile), use it and skip inference.

Otherwise, scan the text for stack signals:

Signals	Stack
Java, Spring, Maven, JPA, Flyway, Testcontainers, Hibernate, @Bean, pom.xml	backend
TypeScript, Next.js, React (web), Vitest, Playwright, Tailwind, next-intl, web/	web
Expo, React Native, Maestro, jest-expo, mobile/, Android, EAS, app.json	mobile
None of the above, or applies across stacks	cross-cutting

---

Step 2: Determine target file and section

Classification	Target file	Target section
backend	backend/CLAUDE.md	## Known Pitfalls
web	web/CLAUDE.md	## Known Pitfalls
mobile	mobile/CLAUDE.md	## Known Issues / Pitfalls
cross-cutting	memory/MEMORY.md	## Key Learnings → pick the closest subsection

---

Step 3: Dedup across ALL files

Extract 2–3 unique key phrases from the text (specific method names, error messages, or unique terms).

Search for these phrases across:

• CLAUDE.md (root)
• backend/CLAUDE.md
• web/CLAUDE.md
• mobile/CLAUDE.md
• memory/MEMORY.md

If an EXACT match is found:

• Report: "Already exists in {file} → {section}. Skipped."
• Stop — do NOT write.

If a NEAR match is found (same topic, different wording):

• Show the existing entry and the new text side by side.
• Ask: "A similar entry exists in {file}. Do you want to: (a) update existing, (b) add as new entry, (c) skip?"
• Proceed based on response.

If no match found: proceed to Step 4.

---

Step 4: Write to target file

Append the pre-written text to the correct section, with (Added YYYY-MM-DD) timestamp if not already present in the text.

For MEMORY.md Key Learnings: place under the correct subsection (Mobile, Infrastructure, Flyway, JaCoCo, Spring/Testing, Backend, Playwright, Frontend, Python, Git, etc.). Create a new subsection only if none of the existing ones fit.

Remember: MEMORY.md Key Learnings carries the comment:

<!-- Stack-specific pitfalls belong in {stack}/CLAUDE.md. Only keep entries here that are NOT in any stack file. -->

Verify classification before writing to MEMORY.md — cross-cutting entries only.

---

Step 5: Report

Output a single confirmation line:

Saved to backend/CLAUDE.md → Known Pitfalls

or

Skipped — already exists in web/CLAUDE.md → Known Pitfalls (line ~175)