prepare-automation-knowledgelisted

# prepare-automation-knowledge Run this at the **planning stage of every automation/project** (after the `route-project-or-pages-request` fork said PROJECT), before anything is built. > **Projects layer manifest.** Fractera agents do not run an automation once on request — they > build a platform for developing **repeatable** automations: standardized reuse, a visual > interface, input and result data in the local DB and vector memory, quick switching from the UI. > Hermes plus a coding agent build a finished-cycle tool — an **"n8n for one single task"**: the > owner does not recreate the task, they open it in the UI, run it and track the result. ## The three planning truths 1. **The world has ready tools — name them in the plan.** An automation almost never starts from zero. Explicitly consider, per project: web search services (e.g. **exa.ai**), platform APIs (e.g. the **YouTube API**, Telegram, RSS feeds), and **ready-made skills / MCP connectors published on the internet** — search for an existing one before anyone builds a new one. Every chosen integration becomes an env key materialized ONLY via the `persist-env-var-with-rebuild` channel (setter → `app/.env.local` → rebuild). Never hardcode keys. 2. **Delegation is the default.** The planner (Hermes or any orchestrating agent) works in tandem with a usually MORE CAPABLE coding agent. Instead of building the automation yourself, hand it off with the **fullest possible specification file**: the go