use-deep-memory
Skill61 starsChanged 55 days agoMITView on GitHub
Work skill. The person wants something kept for later or brought back — «запомни», «запиши», «сохрани», «не забудь», «напомни о», «что я говорил про…», /remember or /remind from the bot's menu, in any language and any words — or sends a photo, a document or a voice note with such words in the same or the next message. Load it before npm run remember or npm run remind.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- Fractera/Agentic-Engineering-Infrastructureword for word, first seen Nov 8, 2026
Other versions1 project uses a version of this.
Word for word · 1
- Fractera/Agentic-Engineering-Infrastructure★ 61100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository88 other files it gives its agents.
- app/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adopt-suggested-skillSkill
- ai-elementsSkill
- ai-sdkSkill
- answer-elementSkill
- ask-elementSkill
- audit-broken-charactersSkill
- blocks-of-the-nodeSkill
- build-product-with-ownerSkill
- content-factorySkill
- core-build-returnSkill
- core-scheduleSkill
- create-elementSkill
- create-multilingual-content-entrySkill
- design-and-settings-of-elementsSkill
- element-powerSkill
- expand-site-languageSkill
- explain-this-projectSkill
- extract-design-systemSkill
- find-skillsSkill
- handle-block-requestSkill
- handle-needSkill
- help-in-browserSkill
- humanizerSkill
- integratorSkill
- know-the-architectSkill
- manage-app-settingsSkill
- node-typeSkill
- own-agentsSkill
- persist-env-var-with-rebuildSkill
- place-page-in-menuSkill
- provider-rulesSkill
- release-templateSkill
- review-pageSkill
- route-to-elementSkill
- self-reviewSkill
- shadcnSkill
- shape-the-needSkill
- skill-creatorSkill
- study-for-elementSkill
- supervise-elementSkill
- translate-pendingSkill
- ultracode-sessionSkill
- use-agentic-ragSkill
- use-ai-generationSkill
- use-aioSkill
- use-app-configSkill
- use-auth-providersSkill
- use-authSkill
- use-browserSkill
- use-channelsSkill
- use-chatSkill
- use-code-shapeSkill
- use-databaseSkill
- use-dataSkill
- use-deploySkill
- use-designSkill
- use-development-stepsSkill
- use-domainSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 15
Read what the person wants:
- Distinctive: line 60
Every element keeps its own space in the memory and in every store behind it, opened with its own key (node step 654): an element never …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
The node's deep memory: remember and remind
Process: working — data only: no commit, no build. Answer in a few lines, in the person's language.
A hint, not a law: models of late 2026 work worse under rigid step-by-step instructions.
A memory request — by purpose, never by a word
The owner, 2026-10-26: «анализ прямого совпадения это вариант который нельзя использовать» — 82 languages, typos, a phone's autocorrect. Read what the person wants:
- to keep a fact, a thing or a document for later → remember;
- to get back what was said or kept → remind;
- how you should work with them («отвечай коротко», «зови меня Ромой») → your picture of them (
know-the-architect), and remember too when they asked to remember it; - the same word in another task is not memory: «запиши меня к врачу» (a booking), «сохрани файл в проект» (a build), «пометь задачу
готовой», «напомни завтра в 9 позвонить маме» (a time — a schedule,
core-schedule); - a decision about the product or the work you are shaping with the person right now («запомни, у каждой соцсети десятки аккаунтов»
in a talk about their projects) is a decision of the step: verbatim into
current-steps.md(skilluse-development-steps), not the deep memory (node step 683; ✗ 2026-11-07 the owner: «не уверен что мы должны через память их отслеживать»). Unclear — one short question.
A thing and its words — one call
Telegram sends a photo, a document or a voice note as a message of its own; its words most often come in the next one («фото» → «запомни, это моя мама»), sometimes before («сейчас пришлю фото мамы, запомни»); several photos may share one phrase. The memory ties a thing to its words only inside one call, so you are the one who pairs them:
- a photo arrives with
image_path— that file; a document, a voice note, audio or video arrive withattachment_file_id— calldownload_attachmentfirst and take the path it returns; a message the node passed while you were off says «[вложение: …, file_id …]» — the same; - a thing with no words yet → do not store it; one line: «Получил фото. Что с ним сделать? Например: «запомни, это моя мама»». The words that come next are its words;
- words that announce a thing → wait for the thing, then one call;
- then
npm run remember -- "<the person's words, as said>" --file <path> [--file <path> …].
Remember
npm run remember -- "<the words, as said>" [--file …]. The command turns the memory element on for the task and off after (the first start
≈ 20 s, then seconds; a phrase ≈ 15–30 s, a thing longer). In Telegram reply first, in one line: «Принято, запоминаю — около минуты»; then the
result. Say what memory kept, from its answer, in plain words — never more than it said. A refusal — its reason, in the person's words.
Remind
npm run remind -- "<the question>" prints two sources: the deep memory (element memory — facts, things, the date, how sure) and the
diary of everything said in the node (found by meaning). Give one answer: the fact, its date and where it comes from («из глубокой памяти»,
«из нашего разговора 24 октября»); a guess said as a guess. A found thing prints its file path — attach it to your Telegram reply (the reply
tool's files). Nothing in either — «об этом ничего не записано».
How it looks in Telegram (the owner, 2026-11-03: «Цитата + значок + шапка (Рекомендую)»): the reply tool with format:
"markdownv2" — the first line is the mark and the header in bold, the words a quote, one > per line. In the text put a backslash before each of
_ * [ ] ( ) ~ > # + - = | { } . ! and the backquote; a send refused for its entities goes again as plain text. The deep memory's answer: 🟢 *Глубокая память*, then the answer; reminders («⏰ Напоминание · …») come by
themselves and are never written by you.
Whose memory
Every element keeps its own space in the memory and in every store behind it, opened with its own key (node step 654): an element never
finds what was said through another. You, the main agent, read every space with the node's data key and say whose each finding is;
what you keep yourself goes into the space node.
Never
- Store or recall when the person switched memory off («не запоминай»,
know-the-architect): the commands refuse until «запоминай снова». - Turn the memory element on by hand for these — the commands do it.
- Promise forgetting: the deep memory cannot forget yet (its graph keeps what it learned); web pages and YouTube links are not read on this node; a video needs ffmpeg on the machine.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

