Extend intent-gap patch-restore re-drive to the sweep workflow

[pbean]

Summary

The intent-gap patch-restore re-drive (added for story runs) should be extended to the sweep workflow (SweepEngine deferred-work bundles).

Background

BMAD-METHOD PR #2564 ("preserve the attempted change on intent-gap halts") makes bmad-dev-auto save the reverted attempt as a patch file before halting on an intent gap at the review step. The orchestrator now has a sanctioned recovery for story escalations: bmad-loop resolve (interactive, or --no-interactive --restore-patch <path>) re-arms the escalated spec to in-review, latches the patch onto the task, and the engine re-applies it onto the baseline after every reset of the re-drive — so the re-driven dev session resumes review on the restored diff instead of re-implementing from scratch.

This landed on branch feat/intent-gap-patch-restore (plan: i-received-a-message-unified-journal.md). Scope was deliberately limited to story runs; sweep-bundle escalations still take a from-scratch re-drive.

What's missing

SweepEngine bundle escalations should honor a restore_patch latch the same way story runs do:

• re-arm the bundle spec to in-review (mirroring runs.rearm_escalation's restore branch), rather than the bundle's from-scratch equivalent;
• re-apply the saved patch after the bundle's baseline reset, before dispatching the bundle dev session (mirroring Engine._restore_patch, called from the dev loop on feedback is None);
• clear the latch on the bundle commit (mirroring Engine._commit clearing task.restore_patch = None next to resolved_redrive = False).

Note the sweep path has its own _drive_story / _rollback_or_pause / _safe_reset call sites (sweep.py), and bundles have no sprint-status entry, so the story-run wiring can't just be inherited unchanged — the seams need porting.

Design references

• model.StoryTask.restore_patch — the persisted latch (already exists; bundle tasks reuse StoryTask).
• verify.apply_patch(repo, patch_path) — the git apply helper (already exists; reusable as-is).
• Engine._restore_patch — the apply/journal/escalate-on-failure helper and its feedback is None gating in _dev_phase (the pattern to mirror in the sweep dev loop).
• runs.rearm_escalation(..., restore_patch=...) — the in-review vs ready-for-dev re-arm branch (the sweep bundle re-arm needs the equivalent).

The sweep resolve/re-arm entry point (however bundle escalations are resolved) would need a restore_patch passthrough analogous to cli.cmd_resolve.

[coderabbitai]

🔗 Related PRs

#53 - fix(adapters): complete dev sessions only on Stop or window death [merged]
#54 - feat(scm): bounded retention for attempt-preserve-dirty worktree-snapshot refs [merged]
#62 - fix(engine): session-end durability + raw KeyboardInterrupt handling [merged]
#63 - fix(engine): close resume-replay edges from #62 (final-cycle replay, stale result_json, record bloat) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!