Skip to content

v5.9.0

Choose a tag to compare

View all tags
@rhuss rhuss released this 04 Jul 07:12
v5.9.0
3d1c0fd

What's New

Submit/finish workflow split

  • The finish command has been split into submit (creates or updates the PR) and finish (merges and cleans up)
  • This lets you iterate on PR feedback without re-running the full merge pipeline

Review idea inbox

  • Out-of-scope ideas discovered during triage, deep review, and brainstorm are now captured in an inbox file
  • Inbox entries are deduplicated and ordered by relevance
  • New drain mechanics let you process captured ideas when ready

Upstream contribution mode (spex-detach extension)

  • New spex-detach extension creates clean PR branches with spec artifacts stripped
  • Designed for contributing to upstream projects that don't use spec-driven development
  • Opt-in during init, with automatic detection in the finish pipeline

Worktree CWD persistence

  • Default worktree location moved to .claude/worktrees/ so Claude Code automatically persists the working directory across session restarts
  • Statusline now reads worktree-specific state in multi-worktree setups
  • CWD recovery extracted into a dedicated script to prevent resets after subagent runs

Triage hardening

  • CodeRabbit rate-limit detection with automatic fallback to local CLI review
  • Deep-parse Codecov comments for per-file coverage data with bot cross-reference
  • Status bot detection (Codecov, Dependabot, etc.) to separate bot noise from human feedback
  • CI status check added after triage fixes (Step 7b)

Closeout gate for unresolved findings

  • New deterministic quality gate that blocks ship when review findings remain unresolved
  • Surfaces the exact findings that need attention before merge

Refactoring: plugin root references

  • Replaced find-based script resolution with PLUGIN_ROOT from spex-context
  • All relative script paths converted to absolute references for reliability

Fix: pipeline and workflow corrections

  • Post-pipeline completion now uses AskUserQuestion instead of free-form text
  • YOLO mode uses bypassPermissions instead of auto for correct behavior
  • Smoke test removed as a pipeline stage (reduced to 8 stages)
  • Auto-commit question removed from init
  • Successful server-side PR merges detected despite local fast-forward divergence
  • Deep-review skill name corrected from .review to .run
  • Stale .spex-state cleaned from main repo after worktree creation

Full Changelog: v5.8.0...v5.9.0

Assets 2
Loading