Weave v0.6.0

Highlights

Skills via OpenCode SDK — Skills are now loaded through the OpenCode SDK HTTP endpoint instead of filesystem scanning, removing dead discovery code and simplifying the architecture.

Tapestry post-execution review — Post-execution review is now consolidated into Tapestry with git start SHA tracking for accurate cross-session diffs, replacing the previous Loom handoff pattern.

Per-task verification protocol — Tapestry now runs structured self-verification after each task, using Edit/Write tool call history and LSP error detection instead of full build checks.

What's Changed

Added

• Replace filesystem skill scanning with OpenCode SDK integration (8a15dc5)
• Move post-execution review into Tapestry with start SHA tracking for cross-session diffs (4f4216a)
• Automated post-execution review handoff to Loom after Tapestry completes (9bdfb7b)
• Tapestry per-task self-verification protocol (f8d863e)

Fixed

• Resolve config-defined skills in agent overrides (62765f1)
• Tapestry reports review findings instead of autonomously fixing them (99f5a14)
• Direct users to Loom (not Pattern) for plan creation (8ee1518)
• Prevent post-execution review loop by tracking review_triggered flag (5183e04)
• Use display name for Loom in work-continuation handoff (95c2752)
• Route post-execution review to Loom when plan completes (2e4dc5c)
• Fix /start-work command — Loom stays active and delegates to Tapestry (7d76683, 6f823b0, b6f1bac)

Changed

• Streamline Tapestry verification — use Edit/Write tool call history and LSP instead of git diff and build checks
• Remove Tapestry→Loom agent-switch hacks and mandatory post-execution review gate
• Clear state.json on plan completion for faster session.idle exit

Full Changelog: https://github.com/pgermishuys/weave/compare/v0.5.2...v0.6.0