Przejdź do treści
Biblioteka skilli
Zarządzanie i produkt · Automatyzacja

operate

/operate

Pozwala sesji administrować działającą aplikacją Pithy na stagingu lub produkcji, zaczynając od odczytów i pytając przed zmianami.

Repozytorium
pithy ↗
Licencja
MIT
Ostatnia zmiana repozytorium
11.11.2026

Ocena i bezpieczeństwo

95/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść30 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Pozwala sesji administrować działającą aplikacją Pithy na stagingu lub produkcji, zaczynając od odczytów i pytając przed zmianami.

Argumenty

/operate [staging|prod]
Oryginalny opis (EN)

Lets this session administer a deployed Pithy app's staging or production environment, through pithy mcp. Connects an agent with the reads first, asks before every call that changes anything, widens the grant only on a yes. Use when someone wants the session to read the audit trail, triage support, check a user or otherwise operate a live app.

Jak działa

  1. 1. Which environment
  2. 2. Grant the reads first
  3. 3. Call
  4. 4. Widen the grant
  5. 5. Close it
  6. CI
  7. Report
Sekcje z nagłówków SKILL.md, w oryginale · 58 linii, 757 słów
Podgląd SKILL.md
---
name: operate
description: Lets this session administer a deployed Pithy app's staging or production environment, through pithy mcp. Connects an agent with the reads first, asks before every call that changes anything, widens the grant only on a yes. Use when someone wants the session to read the audit trail, triage support, check a user or otherwise operate a live app.
argument-hint: "[staging|prod]"
---

Let this session administer one deployed environment through its own control-plane connection. You are the main session. The tools are the Worker's own admin routes, served by `pithy mcp serve` on this machine, and every call is verified, refused or allowed, and audited by the person's Worker with actor kind `agent`. Nothing is hosted.

This is not the app's own `/mcp` endpoint, where the app's users connect their agents. That one is `mcp({ floor })`, and `/app-design` settles it.

## 1. Which environment

`staging` unless they say otherwise. One connection per environment: a yes for staging is not a yes for production. Check what is already there with `pithy mcp status --json`, which is yours to run.

**An environment holds one connection: the dashboard's or an agent's, never both.** Read `actorKind` from `pithy dashboard status --env <env> --json`, which is yours to run. `connected: false` means nothing is there: go to step 2. `actorKind` `dashboard`, `script` or null means the connect is refused: say so, and leave the choice to the person. Never disconnect it yourself. `agent` means an agent's connection already exists, perhaps a teammate's or this person's on another machine: a new key joins it and takes its grant, so offer the step 2 command without `--scope`, and change the grant only through step 4.

## 2. Grant the reads first

Start with `manifest:read` plus only the `reads` scopes the person needs. `manifest:read` is required: the server reads the manifest under it on every tool list, and without it no tool is offered. Never `--scope all`, and never a write, a send, a charge, a delete or a credential scope on the first grant. Ask what the connection is for, in their words: it is recorded as its purpose.

Offer it as `! pithy mcp connect …`, for the person to run themselves: it registers a credential under their signed-in session. Always name the grant, one `--scope <read>` per read the person needs:

```bash
! pithy mcp connect --env <env> --purpose '<their words>' --client claude-code --scope manifest:read --scope <read>
```

A `!` command has no terminal, so a connect that names no `--scope` asks nobody and stores the default grant: every read plus `keys:rotate`, which mints credentials. Only when the person runs it without `--scope` in their own terminal does it list every operation, preselected to that same default, for them to deselect `keys:rotate` and anything past what they asked for.

Once it reports `Done.`, tell them to restart Claude Code, or run `/mcp`, so the session picks up the new server. Its tools are named `<capability>_<operation>`.

## 3. Call

A read needs no question. Read each tool's kind from its MCP annotations, never from its name:

- `readOnlyHint: true` is a read. Call it.
- Every tool carries `readOnlyHint`, so anything with `readOnlyHint: false` needs a yes: it writes, sends, charges, deletes or changes a credential. `destructiveHint: true` marks a charge, a delete or a credential change; the description's first sentence says which.

**Before any call that is not a read, ask for a yes for that call.** Name the tool, what it will do in the description's words, and its input. A yes is for one call: never for the next, and never for a loop.

A call the Worker refuses for a missing scope is not offered again under another tool. Say what it needs and go to step 4.

## 4. Widen the grant

Only on the person's yes, naming each scope and its kind. `--update` replaces the grant with exactly the scopes given, so pass every scope to keep, `manifest:read` included:

```bash
! pithy mcp connect --env <env> --update --scope manifest:read --scope <scope>
```

Then `/mcp` again.

## 5. Close it

`pithy mcp disconnect` is the person's. It revokes the key on their Worker and removes the key file, the client entry and the record. Give them the command, `pithy mcp disconnect --connection <alias>`, with the alias `pithy mcp status --json` names.

## CI

A job connects with a workload trust instead of a key, through `pithy mcp connect --env <env> --purpose '<text>' --oidc-issuer <issuer> --oidc-claim <name=value> --scope manifest:read --scope <read>`, and holds no key at all. Name the grant here too: away from a terminal, a trust registered with no `--scope` holds the same default as step 2's. It is the person's to set up.

## Report

The environment, the connection's alias and grant, each call that changed anything and the yes it ran on, and anything left for the person: a scope to widen, a connection to close.

Wersje

RepozytoriumWynikLicencja
Ta strona · pithy-sh/pithy95MIT
vecten/vecten-open100Apache-2.0
ndif-team/skills100MIT
kamp-us/phoenix91MIT
nazmirket/skills90MIT

Migawka bazy z 13.11.2026 · wersja paczki 8007271be01a (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 90deploy
    Wdraża aplikację Pithy na staging lub produkcję i podłącza dashboard, po sprawdzeniu danych dostępowych i potwierdzeniu kroków.
  • 100new-app
    Zakłada nową aplikację na platformie Pithy: pyta o web lub mobile, tworzy projekt poleceniem pithy init, dodaje funkcje i dołącza tylko potrzebnych agentów.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.