Spawner UI is the local execution plane for Spark missions. It should be treated as a local control surface, not a public internet service.

• Does not own Telegram ingress.
• Does not receive Telegram webhooks.
• Receives mission requests from spark-telegram-bot and local UI flows.
• Sends mission lifecycle events back to Telegram through the secret-protected local relay.

Never commit:

• .env, .env.*
• provider API keys
• Telegram bot tokens
• local logs, caches, mission state, or generated project files containing private data

Spark CLI starter setup should provide MISSION_CONTROL_WEBHOOK_URLS and TELEGRAM_RELAY_SECRET. Keep cloud LLM API keys in the Spark secret backend where possible; Spawner should receive non-secret provider metadata unless a provider integration explicitly needs a key at runtime.

Run:

npm run check
npm run test:run
npm run build
npm run smoke:routes
npm run smoke:mission-surfaces
npm audit --omit=dev --audit-level=high

Do not expose local control APIs on 0.0.0.0 without an explicit auth and origin policy.

The Spark agent bridge (/api/spark-agent/*) is local-first. If it is exposed outside loopback, configure:

• SPARK_AGENT_API_KEY
• SPARK_AGENT_ALLOWED_ORIGINS

Session commands also support x-spark-agent-session-id scoping. Keep this aligned with docs/SPARK_AGENT_BRIDGE_API.md.

/api/access/execution-lanes is a read-only capability endpoint. It redacts local filesystem paths for unauthenticated browser callers; configured UI, bridge, or MCP API credentials can receive full local workspace details.

Leave SPARK_DOCKER_AVAILABLE unset for normal installs so Spawner can probe docker info with a short timeout and cache. Set it to 1 or 0 only when an install needs an explicit Docker availability override. Keep SPARK_CODEX_SANDBOX=workspace-write and SPARK_ALLOW_HIGH_AGENCY_WORKERS=0 unless this is a trusted local operator install.