Fix exec-harness automation reliability gaps #31

Objective

Fix the known exec-harness automation gaps before relying on the harness as the regression gate for Codex Lab-specific behavior.

Finish Line

Known exec-harness automation gaps are fixed and covered by focused validation.

Current Status

State: Active. Next action: create a task branch and reproduce the just local-cleanup-space --apply argument forwarding failure.

Scope

In:

• Fix just local-cleanup-space --apply argument forwarding.
• Update .github/workflows/exec-harness.yml path filters to include local helper scripts.
• Add focused validation for dry-run versus apply behavior where practical.

Out:

• No broad harness refactor.
• No feature scenario work beyond what is needed to prove the automation fixes.

Acceptance Criteria

• just local-cleanup-space remains a dry run.
• just local-cleanup-space --apply forwards --apply to scripts/local/cleanup-space.sh.
• Exec-harness workflow triggers when scripts/local/exec-harness-env.sh or scripts/local/cleanup-space.sh changes.
• PR validation records the exact commands or CI evidence used.

Relationships

Child of #28. This should be completed before new harness scenarios become required gates. Related to #22 for broader fork CI posture.

Validation

Run the focused shell/script tests added for this change, plus any existing local script test target that owns these helpers.

Decisions

This is the first implementation slice after plan review because broken automation can silently invalidate later regression work.

Open Questions

Should scripts/local/** be the path filter, or should the workflow enumerate only the two known helper scripts?