Fix concurrent-orchestrate noise, stash false-positives, and stale-cache workaround

[jodavis-claude]

Summary

• concurrent_schedule.py now blocks internally via poll_until_actionable(), polling immediately and then pausing 30s between up to 10 idle cycles (5 minutes, both configurable) before returning control — instead of the calling agent re-invoking it every 30 seconds. concurrent-orchestrate/SKILL.md updated to match (longer Bash timeout, simplified wait/re-invoke logic).
• ensure-working-branch and watch-pr's worktree-freshness checks no longer check git stash list — the stash is shared across every worktree in a repo, so a stash left by unrelated concurrent work was tripping an unrelated task's hard-stop. git status --short (worktree-local) is kept.
• workflow-orchestrate/SKILL.md now explicitly requires verbose, detailed failure reporting for any termination outside the normal "done" path (unhandled exception, unrecognized descriptor shape, uncovered tool failure) — terseness is only for the routine success/failure case. Added an explicit fallback branch for an unrecognized descriptor shape.
• Bumped plugins/dev-team/.claude-plugin/plugin.json to 1.4.0. The installed plugin cache was pinned to a pre-ADR-314 commit that still had the removed todo-log/TodoWrite-mirroring plumbing, and hadn't been bumped in a long time so Claude Code never re-synced it.

Test plan

• python3 -m pytest plugins/dev-team/skills/concurrent-orchestrate/scripts/test_concurrent_schedule.py -q — 23 passed, including new coverage for poll_until_actionable's immediate-return, sleep-and-retry, and give-up-after-budget cases.
• Verified the refreshed plugin cache (claude plugin update) no longer contains any todo_log/TodoWrite references.

🤖 Generated with Claude Code

https://claude.ai/code/session_011cC89FQUS5DpmfcnmxrLew

[github-actions]

build-and-test: Python test results

Status: ✅ Passed

Test log