phase-loop

Bridges harness TUIs to the repo-local phase-loop runner for bounded roadmap execution.

2|1|Updated Jun 25, 2026
One-click install
npx skills add https://github.com/Consiliency/agent-harness --skill phase-loop-consiliency
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: phase-loop
Source: https://github.com/Consiliency/agent-harness/tree/main/phase-loop-skills/phase-loop
Command: npx skills add https://github.com/Consiliency/agent-harness --skill phase-loop-consiliency

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Operators working inside an AI harness TUI (Claude, Gemini, OpenCode, Codex) need a safe, consistent way to check status, resume, and run bounded phases of a roadmap-driven automation loop without reimplementing the runner's reconciliation, handoff, or execution logic. ## Core Features & Use Cases - Bounded loop control: Run, dry-run, resume, status, and monitor commands against the shared .phase-loop/ state, with interactive runs clamped to --max-phases 1 by default. - Blocker and closeout handling: Preserves the frozen human-required blocker taxonomy, spec-delta closeout decisions, and artifact-backed re-verdicting so gates only change verdict via runner-owned evidence. - Skill maintenance and drift repair: Audits or repairs installed *-phase-loop bridge skills via sync-skills, and delegates skill-maintenance workflows to the runner. - Use Case: An operator returns to a TUI session after another control plane drove the loop; they invoke the handoff command to read .phase-loop/tui-handoff.md, verify state with status --json, then launch a bounded repair or next-phase run. ## Quick Start Ask the assistant to run the phase-loop status command to report the current reconciled loop state from the shared .phase-loop state files.

Frequently Asked Questions about phase-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a bounded phase-loop execution from a TUI?▼

Invoke the run command with an explicit bound, such as phase-loop run --max-phases 1. The bridge clamps unbounded interactive requests to a single phase and reports that TUI-initiated runs stay bounded.

How do I resume a phase-loop after another session drove it?▼

Start with the handoff command or read .phase-loop/tui-handoff.md, then verify exact machine state with state --json or status --json before resuming. The handoff links to state, events, runs, and the latest log.

Does the phase-loop bridge work with Claude, Gemini, and OpenCode?▼

Yes, per-harness override SKILL.md files adapt the same bridge contract for Claude, Gemini, OpenCode, and Codex. Each override maps to the shared runner commands while preserving harness-specific launch posture and trust metadata.

What happens when the phase-loop reports a human-required blocker?▼

The bridge preserves the blocker_class, summary, required inputs, and automation status from the frozen taxonomy, such as dirty_worktree_conflict or branch_sync_conflict. For access blockers it performs metadata-only probes and never prints secret values.

Why does a phase-loop run fail with branch_sync_conflict?▼

In pipeline mode with branch governance on, dispatch switches to a branch cut from origin/main, so a roadmap committed only on an unpushed local branch is refused cleanly. Push the roadmap to the base branch or pass --allow-branchgov to switch anyway.

When should I use sync-skills on the phase-loop bridge?▼

Run sync-skills --check when the advisory bridge record reports missing_root, missing_skill, or drifted for installed bridge skills. Use --apply only when you explicitly want to repair local bridge-skill drift for manual reentry.