Verify the replayed branch from Go #898

Problem

Verification at packages/agents/source/skills/rebase/shared/scripts/verify-rebase.sh:56 chooses the old tip, compares explicit ranges, flags dropped work and consumes evidence only when no problem remains. Porting only range-diff would lose those recovery rules.

Part of #588 and #326.

Acceptance criteria

1. AC1: Given an active rebase, when verification is requested, then it refuses until the replay finishes.
2. AC2: Given an explicit old tip, completed evidence or only ORIG_HEAD, when verification selects the comparison, then it follows the frozen precedence.
3. AC3: Given incomplete evidence and no explicit old tip, when verification runs, then it refuses after removing that stale evidence.
4. AC4: Given valid saved evidence, when the replay is compared, then its old and new ranges match the frozen explicit-range selection.
5. AC5: Given missing evidence, empty ranges, empty comparison output or dropped work, when verification reports, then its attention status matches the frozen implementation.
6. AC6: Given comparison output above the configured cap, when verification reports, then it prints the frozen pairing summary.
7. AC7: Given committed-tree marker candidates, when verification reports, then its finding matches the frozen candidate, content and exemption behavior.
8. AC8: Given a changed path declared for regeneration, when verification reports, then it flags that path for regeneration.
9. AC9: Given dirty state and gate availability, when verification reports, then it prints the frozen working-tree and gate advice without running gates.
10. AC10: Given no attention finding, when verification completes, then completed replay evidence is consumed.
11. AC11: Given an attention finding after valid completed evidence, when verification completes, then that evidence remains available.
12. AC12: Non-functional: native verification cases execute without skips on macOS, Linux and Windows.

Scope and non-goals

Scope: post-rebase reporting and evidence consumption.

Non-goal: running gates, regenerating, committing fixes, or correcting the historical marker candidate/content distinction. Keep shell and coupled Python frozen through N-2, with no new shell logic. Do not switch callers or retire tests in this child.

Approach

Port range and empty-range handling from packages/agents/source/skills/rebase/shared/scripts/verify-rebase.sh:114. Preserve the current marker behavior explicitly: packages/agents/source/skills/rebase/shared/scripts/verify-rebase.sh:183 selects candidates from HEAD, then reads their worktree contents. Do not describe that as scanning committed blobs or silently change it. Pin a differing-HEAD/worktree fixture. Reuse child 01 and the Go scanner. Writes are confined to established evidence cleanup.

Dependencies

Blocked by: #895, #753.

Child 02 supplies completed-replay fixtures before activation.

Size

M. Forecast 700 code lines: 300 implementation, 340 tests and fixtures, 60 integration. Measured source baseline: 184 code lines. Prose allowance: 220 lines.

scc could not run in the isolated checkout. Measured baselines count non-blank lines other than full-line # comments, including Python docstrings and embedded code conservatively. Forecasts include implementation, tests, fixtures and registration; they are estimates, not measured future Go. Generated output is excluded only under packages/agents/claude/.apm/skills/, packages/agents/codex/.apm/skills/, .claude/skills/, .agents/skills/ and tokens.json. The source under packages/agents/source/ and copies anywhere else count. The final diff must be at most 800 code lines and 2,000 prose lines, with no exception; split before implementation if either forecast grows beyond its limit.

Verify plan

• AC1, AC2, AC3: compare active, explicit-tip, completed-record, aborted-record and ORIG_HEAD fixtures.
• AC4, AC5, AC6: compare ranges and reports for no merge base, zero old/new commits, dropped work, empty output and cap boundaries.
• AC7: compare committed-marker/worktree-clean and committed-clean/worktree-marker cases, exemptions and scanner failure with frozen behavior.
• AC8, AC9: exercise attributes, dirty trees and task probes; assert no generator or gate invocation.
• AC10, AC11: inspect evidence after clean and attention results; verify index and contents stay unchanged.
• AC12: inspect executing native cases on all three CI platforms.
• Definition of done: D1 each AC is verified at the merged head, with the commands and results cited in the pull request's Verification. D2 an independent review approves the exact merged head, and every finding has a recorded disposition. D3 required checks pass with no gate loosened. D4 the leak scans the repository specifies are clean, and every post made for this work passes the per-post leak review before it is posted. D5 every commit is signed. D6 the documentation matches the change or the Verification says why there is no impact. D7 every commit has an Assisted-by trailer and the pull request has the provenance:agent-assisted label. D8 the pull request stays within the size limits. D9 follow-ups are filed as issues and listed under Related. D10 the merge closes the issue through a closing reference, the board shows Done, and the final Status: comment names the merge commit.

Risks

A wrong comparison can hide dropped work; early cleanup destroys evidence. The marker candidate/content distinction stays a documented limitation. Undo: revert this child and retain investigation evidence.

Operator authority

None.