Skip to content

Latest commit

 

History

History
98 lines (77 loc) · 4.28 KB

File metadata and controls

98 lines (77 loc) · 4.28 KB

Spark CLI Production Prep Tasks

Last updated: 2026-05-08 Branch: codex/fix-railway-smoke-remote-failures-20260508 Push status: prepared locally; do not push until paired Spark repo updates are ready.

Current State

Spark CLI is through the main Railway/VPS/remote-sandbox hardening work. The remaining work is release hygiene and cross-repo coordination, not a new feature build-out.

Verified locally:

  • python -m pytest -> 497 passed, 5 skipped
  • python -m spark_cli.cli verify --installers --json -> OK
  • python -m spark_cli.cli verify --installers --hosted-installers --json -> OK
  • python -m spark_cli.cli verify --sandboxes --json -> OK
  • python -m spark_cli.cli support bundle --json -> no raw local path leak found
  • git diff --check -> OK

Expected local-only limitation:

  • verify --hosted --json and live verify --json can fail on this desktop when no hosted LLM provider/API keys are configured. That is an environment readiness signal, not a Spark CLI code blocker.

Completed Phases

  • Read user feedback and convert it into a production hardening scope.
  • Hide paused Spark Pro connection-token and bearer-token surfaces from user-facing docs and diagnostics.
  • Document Railway/VPS hosted Spark Live deployment and verification paths.
  • Add secure SSH sandbox docs and CLI readiness checks.
  • Add secure Modal sandbox docs and no-secret smoke checks.
  • Align OWASP/agentic security docs with the shipped sandbox surfaces.
  • Add launch runbook release gates for installers, sandboxes, and hosted Spark Live.
  • Harden SSH/Modal diagnostics against secrets, URL credentials, private key paths, bearer tokens, and local audit paths.
  • Harden public diagnostics, support bundle output, and installer provenance against raw local path leaks.
  • Verify installer, sandbox, support bundle, and test-suite gates locally.
  • Prepare agent-facing safe sandbox guidance and future installer option docs without advertising them as shipped installer features.
  • Prepare detailed sandbox test runbook and evidence template for the May 9, 2026 SSH, Modal, Railway/VPS, and Telegram smoke session.

Remaining Phases

Phase 1: Final Cosmetic Security Normalization

Optional but recommended before the bundled push.

  • Normalize generic repair hints such as ~/.spark/config/secrets.local.json to <spark-home>/config/... in shareable JSON output.
  • Re-run focused diagnostics after that change.
  • Commit the normalization if changed.

Phase 2: Cross-Repo Release Bundle

Required before production push.

  • Wait for the paired Spark repo updates the user wants to ship together.
  • Re-run the Spark CLI launch gate from docs/LAUNCH_RUNBOOK.md on the current prepared branch.
  • Run the hosted installer gate against https://agent.sparkswarm.ai.
  • Run a real Railway/VPS smoke when production credentials are available.
  • Re-run the launch gate again after paired Spark repo updates are landed into the final push bundle.
  • Push only after the full Spark update set is ready.

Latest Phase 2 prep check:

  • python -m pytest -> 497 passed, 5 skipped
  • python -m spark_cli.cli verify --installers --json -> OK
  • python -m spark_cli.cli verify --installers --hosted-installers --json -> OK
  • python -m spark_cli.cli verify --sandboxes --json -> OK
  • python -m spark_cli.cli verify --registry-pins --json -> OK
  • python -m spark_cli.cli verify --provenance --json -> OK
  • git diff --check -> OK
  • safe sandbox agent guidance and future installer option docs are prepared
  • detailed sandbox test runbook and evidence template are prepared
  • python -m spark_cli.cli verify --hosted --json -> expected local failure: no hosted LLM provider configured in this desktop environment
  • python -m spark_cli.cli live verify --json -> expected local failure: hosted mission smoke needs production API keys

Deferred On Purpose

These are not blockers for this release and should not be advertised as shipped:

  • SSH prepare/deploy, remote log tailing, and arbitrary remote shell.
  • Modal arbitrary run, artifact pull, persistent volumes, and provider-secret passthrough.
  • Spark Pro connection tokens and bearer-token entitlement flow.
  • Public inbound hosted services beyond the reviewed Spark Live lane.