Agent Skill
2/7/2026

skill-extract-pattern

Extract patterns from coding agent sessions to improve existing skills or create new ones. Use after completing a work session where you used or needed a skill, when you discovered a new pattern, or when you want to capture workflow knowledge. Works with Claude Code, Pi, and Codex session files. Triggers on "extract pattern", "improve skill from session", "create skill from session", "analyze session".

A
alchimie
1GitHub Stars
1Views
npx skills add alchimie-di-circe/Extractor-desktop-app

SKILL.md

Nameskill-extract-pattern
DescriptionExtract patterns from coding agent sessions to improve existing skills or create new ones. Use after completing a work session where you used or needed a skill, when you discovered a new pattern, or when you want to capture workflow knowledge. Works with Claude Code, Pi, and Codex session files. Triggers on "extract pattern", "improve skill from session", "create skill from session", "analyze session".

Extractor Desktop App (Trae Extractor)

Local-first AI command center for content creators. Extract photos from Apple Photos, edit with AI, generate captions, and schedule posts without uploading personal media to external servers by default.

Visione

  • Automatizzare l'80% del workflow media con agenti AI e controllo umano nei punti critici.
  • Mantenere la privacy: dati locali, API key in keychain, processi isolati.
  • UI desktop reattiva e leggera con Svelte + shadcn-svelte.

Stato e roadmap

La pianificazione e il tracking sono gestiti in .taskmaster/. Vedi:

  • /.taskmaster/docs/prd.md per la visione completa.
  • /.taskmaster/tasks/tasks.json per l'elenco aggiornato delle task.

Focus attuale (estratto dalle task principali):

  • Base UI con shadcn-svelte + routing hash.
  • IPC bridge sicuro + keychain.
  • Configurazione multi-provider LLM.
  • Sidecar Python (FastAPI) per agenti.
  • Agenti: extraction (osxphotos), editing (Cloudinary + Shotstack), captioning (RAG), scheduling (Postiz).
  • Generative UI con json-render (core) + renderer Svelte custom.

Cloudinary MVP Policy (Canonical)

  • Baseline economica: Cloudinary Free con upgrade successivo a Plus quando necessario.
  • Video timeline editor primario: Shotstack.
  • Cloudinary copre upload/tagging/transformation/export e quick edits.
  • Guardrail runtime: preflight centralizzato per limiti piano/quota/add-on e size policy.
  • Storage canonico locale: raw/ (originali) + ready/ (finali pubblicabili).
  • Retention cloud: nessun purge per asset scheduled; purge post publish/cancel.
  • Source of truth: /.taskmaster/docs/cloudinary/README.md.

Stack principale

  • Electron 36 + Electron Forge (main/renderer separati, context isolation).
  • SvelteKit 2 + Svelte 5 Runes.
  • shadcn-svelte + TailwindCSS 4.
  • Task Master AI per PRD e task graph.
  • Python sidecar (FastAPI) per agenti e strumenti AI.
  • Cagent by Docker per definizione agenti e orchestrazione via YAML (non richiede Docker attivo).

Architettura (high level)

Renderer (Svelte UI)
  -> services (IPC client)
    -> Electron preload (contextBridge)
      -> Electron main (IPC handlers, keychain, sidecar manager)
        -> Python sidecar (FastAPI, agenti Cagent)
          -> osxphotos sandbox / MCP tools / Postiz API

Principi chiave:

  • Renderer solo UI, niente I/O pesante o logica di business.
  • Operazioni costose in main process o sidecar.
  • IPC tipizzato e minimale.

Struttura progetto (attuale + pianificata)

electron/                   # Electron main + preload
src/                        # SvelteKit renderer
  lib/components/ui/        # shadcn-svelte
  lib/components/custom/    # componenti app
  lib/services/             # wrapper IPC
  lib/stores/               # Svelte 5 runes
.taskmaster/                # PRD + task graph
python/                     # sidecar agenti FastAPI

Sviluppo locale

Requisiti: Node.js + pnpm + Python 3.11.

Installazione:

pnpm install
python3 -m pip install -r python/requirements-dev.txt

Dev server Electron:

pnpm run dev

Note: On macOS, pnpm run dev is preferred over pnpm run start due to a process signal issue with electron-forge.

Check e test:

pnpm run lint
pnpm run lint:py
pnpm run check
pnpm run test:unit -- --run
pnpm run test:python:smoke
pnpm run pre-pr:checks

Workflow pre-PR (corrente)

Policy sintetica:

  • Gate automatici baseline: changed-files quality (lint-staged --diff), check, test:unit, test:python:smoke quando impatta Python.
  • E2E locale obbligatoria sui flow toccati: Agent Browser CLI (manuale, con checklist nel pre-PR script).
  • Verifica build/package fuori dal gate standard: eseguirla solo per task che toccano electron/main, preload, packaging/release.

Stato E2E Playwright in PR/CI

  • Definizione workflow Playwright in PR/CI è deferred a una sessione dedicata.
  • In questa wave non è stato definito il runner mode CI (Playwright puro vs agent).

Riferimento canonico: /.taskmaster/docs/playwright-e2e-workflow.md

Build:

pnpm run package
pnpm run make

Note:

  • Routing hash: usare #/route nei link.
  • Il progetto usa patching di SvelteKit (vedi package.json).
  • La configurazione agenti vive in YAML (Cagent) e viene generata da settings UI.

Task Master (workflow rapido)

task-master list
task-master next
task-master show <id>
task-master set-status --id=<id> --status=done

Cagent (Docker) - note rapide

  • Gli agenti e l'orchestrazione sono definiti in YAML (Cagent).
  • Non serve Docker attivo sulla macchina host per usare Cagent.
  • I tool MCP possono comunque usare Docker MCP Gateway se disponibile.

Contributing

Linee guida per sviluppo e AI agents in AGENTS.md.

Skills Info
Original Name:skill-extract-patternAuthor:alchimie