What problem does it solve? Operators using Claude Code need a consistent, safe way to inspect and drive the repo-local phase-loop orchestration runner without reimplementing its phase selection, reconciliation, or event-writing logic, and without risking unbounded autonomous runs. ## Core Features & Use Cases - Loop State Inspection: Report reconciled loop state, heartbeats, and terminal summaries via status, state --json, and monitor --once --json commands against the shared .phase-loop/ state surface. - Bounded Execution: Run or dry-run the phase loop with enforced bounds, clamping interactive runs to --max-phases 1 so Claude sessions never launch unbounded TUI runs. - Recovery and Repair: Print TUI recovery handoffs, reconcile durable state on resume, and audit or repair the installed bridge skill with sync-skills --check or --apply. - Use Case: After another harness drove the loop overnight, ask Claude for a handoff and status check; the bridge reads .phase-loop/state.json and tui-handoff.md, reports any frozen blocker such as branch_sync_conflict, and resumes with a bounded single-phase run. ## Quick Start Ask Claude to show the current phase-loop status and handoff using the claude-phase-loop bridge before resuming the run.
