Skip to content

v0.3.1 Stable

This is the canonical public release page for the stable patch release that closes the current Phase 5 operational-closure loop without widening Skylattice's runtime scope.

What This Release Proves

  • Skylattice now has repeatable tracked proof for weekly radar safe-validation instead of a single isolated validation note.
  • Human-readable OpenAI provider instructions now come from tracked prompt files, while runtime code keeps enforcement and parsing boundaries explicit.
  • The release-facing validation story is more honest: the no-credential baseline stays public and repeatable, while auth-dependent checks are recorded as explicit opt-in operator steps.

What Works Today

  • skylattice doctor
  • governed task-agent runs with deterministic edit modes, reviewed memory context, and recovery-aware GitHub collaboration sync
  • skylattice memory ... review, search, export, and rollback flows
  • technology-radar runs with GitHub discovery, provider-tagged evidence, and bounded promotion metadata
  • skylattice radar schedule show, render, run, and validate for tracked local scheduling and weekly review
  • read-only auth preflight plus explicit GitHub bridge guidance
  • public proof artifacts and Pages landing pages for cold visitors

Verify It

Why This Stable Patch Matters

v0.3.0 opened the Phase 5 scheduling and provider-neutral foundation. v0.3.1 turns that story into a clearer operational closeout by making the validation loop repeatable, the prompt truth source explicit, and the auth-dependent reporting boundary easier to trust.

Known Limits

  • task execution remains text-native rather than AST-aware
  • meaningful task planning still depends on OPENAI_API_KEY
  • radar discovery still depends on GITHUB_TOKEN
  • schedule integration is Windows-first in this slice and relies on OS-level task registration
  • the project is still early if you need a broad hosted automation platform