agentleFS
FracteraFractera/web4-node-starter-kit/.claude/skills/supervise-element/SKILL.md

supervise-element

Skill68 starsChanged yesterdayMITView on GitHub

Build process. Lead an element's development for the architect when they chose you (supervise main) — its «done» wakes you; read its state, answer its confirming lines or give the next task, report, stop at the end. Load it when you are woken with «Узел: агент элемента … сообщил, что закончил (надзор …)» or the person asks you to lead or stop leading an element.

  • Sends data out

2 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    leave an element's permission question to the person: since 2026-11-09 (the owner: «агент ядра принимает решение о том что делать», core …

    line 59

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Lead an element for the architect

Process: build — you change no code yourself, yet you lead an element's development: its code changes by the answers and tasks you pass to its agent.

A hint, not a law: models of late 2026 work worse under rigid step-by-step instructions.

Why it exists

The architect said (2026-10-19) that the node, «выступая в качестве старшего руководителя разработки», may follow an element's work and pass it new tasks «без личного участия». You do it only for elements whose record says so; the architect can take it back at any time (npm run supervise -- <id> architect).

The language. An element answers in the language you speak with the person — never English by default (the owner 2026-11-08: «на каком языке ты отвечаешь на том языке должны отвечать элементы»). The first npm run task or npm run wake of a conversation carries --lang <code> of that language (ru when the person writes Russian); it is kept for the next ones and changed by the next --lang when the person switches. The task itself stays English: the line about the language is added by the command.

The node's messages carry a head — ⟦node · <kind> · <element> · <HH:MM>⟧ (step 698): supervision-done, job-done, first-session-done, limit-resume, core-back, held-during-build, build-failed. Act by the kind; what each asks — lib/agent-kit/NODE-LINES.md.

One cycle

  1. npm run state -- <id> — its role, who leads, intake, last commits and current-step.md.
  2. Decide by the task in its intake (the person's words) and its current step:
    • it waits for a «yes» on three confirming lines or a plan → it matches the task: npm run task -- <id> "Да, делай."; it does not: one correction in plain words;
    • it asks a question — a product quiz question too → answer it from the architect's words in its task, quoting them: for this work you are its person (its instruction says so), and the quiz item it records carries those words. Only what the task does not say goes to the architect — tell the element it waits for them (✗ 2026-10-20: the main agent forwarded a quiz question the task answered, «only the person confirms», and the element stopped);
    • a neighbour's request in its intake that its role covers → «Да, делай»; one its role does not cover → ask the architect;
    • a substep or step is closed with evidence → «Дальше по плану» or the next part of the task;
    • the task is done → live first, then the report (the owner, 2026-11-13: «ты можешь и должен в любой момент сам это делать если ты в конвейере задачи»; ✗ the same night «Хранилище картинок» was reported done while its live server still ran the birth build): its work committed and its prebuild green → deploy it yourself, curl -X POST localhost:<core>/api/node/deploy -d '{"ids":["<id>"]}', wait for its /api/health builtAt to move, open the page the task named with npm run eyes and look; only then report to the architect with the link he can click, and npm run supervise -- <id> off. A task is never «done» while the person cannot use it.
  3. Report one or two lines to the architect: what the element did, what you answered. npm run person tells where — Telegram when a bot is connected to you, otherwise this terminal.

When the hand cannot finish (686)

npm run where names it: a turn cut off or stopped by a limit. A stop by the limit already has its resume — the clock wakes you at the reset (686-5); then wake it «continue from your record». The element runs on its own person's subscription (689) and that one ran out before the work ended — finish through the same hand on lent fuel: npm run lend -- <id> --cost <% of the node's window you expect> --reason "<the work>"; it refuses when the forecast leaves you less than 15 % (every running lending counted). Then npm run close and npm run task; the end of that session gives the fuel back by itself, npm run lend shows the debt. Only the work YOU handed out — the architect's own work, led by you; the element's person's Telegram stays off during the loan, and their own wishes never run on the owner's subscription, nor one person's two subscriptions around the limits — skill provider-rules.

Never

  • leave an element's permission question to the person: since 2026-11-09 (the owner: «агент ядра принимает решение о том что делать», core step 704) it comes to you as its own line ⟦node · permission⟧ and you decide it — skill element-permission;
  • build, change or test the element's code yourself;
  • go on after the last cycle: the node removes the record, you report and stop.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.