[jessbuhl]

Audit progress currently records a generic improvement decision without identifying the selected report recommendations. This change saves explicit recommendation IDs, retains their audit context across improvement planning, and adds follow-through status and evidence beneath the same report recommendations.

• Validate selections against the reviewed report and preserve the handoff across scope updates and restarts.
• Keep implementation scope approval separate from recommendation selection.
• Preserve existing freshness checks and avoid automatically rerunning audits after task changes.
• Add deterministic report, selection, resume, legacy-state, and stale-input demonstrations.

Implements ASE-1294.

Validation

• Audit and skill-contract regression suite: 660 passed.
• Full repository run: 6,568 passed, 159 skipped, with two failures found while development was in progress. Both were corrected and passed the final pytest --lf -q rerun (2 passed).
• Passed Ruff lint/format, Linux-targeted type checks for Python 3.12 and 3.13, copyright, license, whitespace, and commit sign-off checks.
• Skips cover optional/native runtimes, platform-specific checks, and opt-in live execution. No live assistant/model runs were performed.
• Reproduction instructions and evidence limitations are in tests/scenarios/audit-guidance.md. Demos use synthetic inputs and scripted decisions; they do not establish live assistant behavior.

Draft integration status

Local validation used base d9bc632. Since then, main gained the audit-history workflow, including removal of findings approval. A merge probe against current main reports conflicts in five files: the audit conversation templates, conversation/report fixtures, and audit report/state tests.

Reconcile this implementation with that newer workflow and rerun affected checks before marking ready. The existing validation does not establish compatibility with current main.