oh-my-claudecode v4.14.0: ultragoal port, autopilot/team launch hardening, plugin skill registry

Release Notes

Release with 1 new feature, 11 bug fixes / hardening changes across 12 merged PRs.

Highlights

• feat(ultragoal): port OMX ultragoal to OMC (#2995) — durable multi-goal workflow with omc ultragoal CLI, persisted plan/ledger artifacts under .omc/ultragoal, and Claude Code /goal handoff guidance. Checkpointing fails closed: only the active in-progress microgoal can be completed, failed, or blocked.
• Preserve Claude Code across OMC updates (#2997) — omc update now snapshots the global @anthropic-ai/claude-code install before running and restores it if the OMC update path removes it; users without a prior global Claude Code install are left untouched.
• Keep plugin skill registry concise (#2989) — re-register every bundled plugin skill now that plugin cache sync compacts native skills/*/SKILL.md files into small registry shims, while preserving full on-demand skill bodies under skill-bodies/*/SKILL.md (loaded via a Windows-safe omc-full-body pointer at invocation time).

New Features

• feat(ultragoal): port OMX ultragoal to OMC (#2995) — @probepark

Bug Fixes & Hardening

• Fix team launch fixed worker plans (#3011) — honor explicit N:agent worker specs for fixed/pre-authored team launch plans, add --no-decompose, fail closed when explicit worker count and decomposed scope count disagree, ignore dead tmux-backed stale team state during spawn gating.
• fix(pre-tool-enforcer): cover extra naming slop boundary (#3013) — neutral domain qualifiers like extra/additional (e.g. extraSecretFetch, extraSecrets []extraSecretFetch) no longer trigger the SLOP fallback warning.
• Harden HUD cache stale lock cleanup (#3003) — resolve the HUD cache dir before cleanup, GC stale render locks early, remove only zero-byte HUD temp/error files (preserve non-empty diagnostics), and clean up render locks and stdout temp files on background-refresh EXIT/signal without blocking the statusLine hot path.
• Fix autopilot state cleanup stop-hook loop (#3001) — make persistent-mode autopilot enforcement read phase ?? current_phase ?? "unspecified", clean orphan autopilot routing echoes, prevent state_clear from clearing a live singleton autopilot owned by another session, and write/read autopilot phase/current_phase aliases for compatibility.
• Preserve Claude Code across OMC updates (#2997) — see Highlights.
• Fix doctor warning for synced omc-reference fallback (#2993) — suppress doctor legacy-skill collision warnings only when ~/.claude/skills/omc-reference/SKILL.md byte-for-byte matches the bundled fallback; modified copies and non-contract omc-reference.md legacy files still warn.
• Suppress persistent reinforcement after oversized tool redirects (#2990) — focused Stop classifier requires both a tool-results/*.txt pointer and oversize/redirect wording; reinforcement is suppressed only for the first few consecutive redirect stops, then normal Ralph/todo stall protection resumes.
• Keep plugin skill registry concise (#2989) — see Highlights.
• Fix doctor package version lookup (#2982).
• Fix launch credential mirroring (#2979).
• Fix cancel-ralph skill alias (#2974).

Stats

• 12 PRs merged | 1 new feature | 11 bug fixes / hardening changes

Install / Update

npm install -g oh-my-claude-sisyphus@4.14.0

Or reinstall the plugin:

claude /install-plugin oh-my-claudecode

Full Changelog: v4.13.7...v4.14.0

Contributors

Thank you to all contributors who made this release possible!

@probepark @Yeachan-Heo