What should go into Claude Desktop's memory versus a Skill?

Systematic knowledge like command allow-lists and type-conversion rules belongs in a Skill, while small quirks found during work go into memory. Once the same insight proves useful three times in different situations, it is a candidate for promotion to a Skill.

How does the engineer avoid stale memory?

By asking Claude to list what it currently remembers, then reviewing each item roughly once a month. Correct items stay, obsolete ones are deleted, and vague ones are reworded and re-saved.

Why not place the whole manual in the connected folder?

The engineer says full manuals are large and slow for the AI to search, so instead only self-written excerpts covering the commands and arguments actually used are placed there, which the engineer says improves reference accuracy.

Large Language ModelsAI Coding AssistantsZenn AI/ML3d agoOct 16, 2026, 10:00 JST

Claude Desktop memory plus folder connection for FA coding

Send on LINE

3 Key Points

What happened
An FA engineer using Claude Desktop for PLC programming describes keeping memory entries for reusable pitfalls — array indices require INT, Execute-type FBs must be dropped after Done — and asking Claude explicitly to remember them, while folder connection lets Claude read per-project specs and self-written manual excerpts.
Why it matters
Session resets normally wipe context, so this practice lets the engineer skip restating premises like 1-based arrays and CR line endings on every new chat, apparently saving setup time on repeated FA work.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

What this piece adds is the day-to-day division of labor: Skills fix unchanging conventions, memory accumulates small conclusions discovered while working, and folder connection swaps in the current project's context.

The engineer's rule for moving something from memory to a Skill — three separate situations where the same insight helped — is one concrete answer to the risk that memory becomes a graveyard of outdated facts. Staying aware of what Claude is being told to remember and which folders it can read is presented as the safety counterpart to the convenience.

A typical day is described as connecting a project folder in the morning, relying on Skills and memory already in place, pointing Claude at the spec to generate FB code, reviewing it, and saving new findings to memory. The next week the folder is swapped for another project while the foundation stays, which the engineer describes as the appealing part of using the tool this way.

FAQ

What should go into Claude Desktop's memory versus a Skill?: Systematic knowledge like command allow-lists and type-conversion rules belongs in a Skill, while small quirks found during work go into memory. Once the same insight proves useful three times in different situations, it is a candidate for promotion to a Skill.
How does the engineer avoid stale memory?: By asking Claude to list what it currently remembers, then reviewing each item roughly once a month. Correct items stay, obsolete ones are deleted, and vague ones are reworded and re-saved.
Why not place the whole manual in the connected folder?: The engineer says full manuals are large and slow for the AI to search, so instead only self-written excerpts covering the commands and arguments actually used are placed there, which the engineer says improves reference accuracy.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.