Slice 2: Mock Call, Real CRM — full implementation + audited (52/53 tasks done)

[brettheap]

Slice 2 — Mock Call, Real CRM

Swap the Slice 1 mock CRM for real Dynamics 365 / Dataverse while keeping the call transport and persona mocked. Ships the full Spec Kit workflow (specify → clarify → plan → checklist ×13 → tasks → analyze ×5 → implement) plus a 10-expert swarm code-review pass with full remediation. Branch is fully audited and passes a clean /speckit-analyze (0 CRITICAL / 0 HIGH / 0 MEDIUM / 0 LOW).

What ships

• DataverseWriteBackAdapter behind the unchanged Slice 1 WriteBackAdapter protocol — 4 operations (Phone Call activity, queue-status update, Task, write-back assembly), idempotency-key stamp + pre-query (FR-024), bounded transient retry (FR-023), If-Match/@odata.etag optimistic concurrency on the final PATCH (T045/Pass-1B).
• DataverseQueueLoader behind the unchanged QueueItem contract — ExplicitId + NextReady selectors, deterministic next-ready tie-break, FR-009 empty-queue no-op vs T051 configured_campaign_not_found distinction.
• OAuth2 client-credentials auth (httpx + Microsoft Entra ID) with redacted error formatters (tenant_id never leaks; FR-005 + T047).
• Metadata discovery + per-run verification (FR-001/FR-002): discover-crm CLI command produces config/dataverse_mapping.json; every write-enabled run re-verifies live.
• dry-run vs write-enabled run modes (FR-031, SC-013) — default dry-run produces planned-writeback artifacts with zero CRM writes; --write performs actual writes.
• Resume coordinator (FR-023, SC-014) — run-crm --resume <session-id> replays only missing emit_* operations via idempotency pre-query short-circuit; fresh baseline at resume start for T045 conflict re-detection on pause-window human changes (CHK061).
• T045 mid-run CRM-state conflict detection — captures last_session_id/status/preserve_if_present/@odata.etag baseline at load; raises CrmConflictError on any mismatch (including row-deletion); maps to exit_status="blocked" + block_reason="conflict_detected".
• Default-on transcript redaction layer (FR-028..FR-030) — regex [REDACTED] policy with summary-only retention option.
• FR-019/020 fixture pre-validation (resolves GitHub issue Transport: validate fixture at place_call time to avoid partial session state #2 — closed via T043) — malformed fixtures rejected before any session-state, queue-status, or attempt-count mutation.
• Structured block_reason discriminator (Pass 1C) — eligibility | metadata | conflict_detected | permanent_other on CrmRunReport and ResumeReport.

Spec Kit artifacts (specs/002-mock-call-real-crm/)

• spec.md — 35 FRs, 15 SCs, 6 user stories, with §Clarifications, §Definitions, §Edge Cases, and §Requirement Coverage Notes.
• plan.md, research.md, data-model.md (incl. T050 state-machine spec), quickstart.md (T042 demo runbook).
• 6 module contracts under contracts/ (incl. T049 run-report JSON schema in cli-slice2.md).
• tasks.md — 53 tasks across 9 phases · 52 done · 1 deferred (T052 architectural follow-up).
• 13 requirements-quality checklists under checklists/ — all 100% closed including reverification.md (72/72).

Implementation tally

Phase	Tasks	Status
1 Setup	T001–T004	✅
2 Foundational	T005–T016	✅
3 US1 MVP write-enabled	T017–T023, T048	✅
4 US2 dry-run	T024–T027	✅
5 US3 metadata block	T028, T029a, T029b, T030, T051	✅
6 US4 idempotency + resume + T045 conflict	T031–T034, T045, T046	✅
7 US5 fixture validation	T035–T036	✅
8 US6 redaction	T037–T039	✅
9 Polish	T040–T044, T047, T049, T050, T051, T052 deferred	✅

Code-review + remediation history (5 analyze passes + 1 swarm review)

Audit	Findings	Outcome
Post-analyze (2026-05-22)	0/0/1/5	Resolved inline
Re-analyze	0/0/1/5	Resolved inline
Post-completion (2026-05-24)	0/0/1/5	T051 + I2–I6 polish
Swarm code-review (10 experts)	0/9/26/25	Pass 1A–1D + Pass 2A–2C + Pass 3 (8 commits)
Post-remediation	0/0/1/3	Pass 4 (1 commit)
Final	0/0/0/0 ✓	Ship-ready

Test + lint

• 763 tests pass (was 432 at branch start)
• 2 pre-existing test_constitution_sync.py failures predate this branch and are out of Slice 2 scope (constitution-template-vs-test format drift)
• ruff check + format: clean
• Adapter coverage: 94% (combined unit + contract + integration); boundary test (T040) verifies SC-010 across 17 modules × 17 patterns = 290 parametrized tests

How to review

This PR is large by line count (~15,800 net additions across 99 files) because it ships a complete vertical slice + spec artifacts + 13 checklists + extensive tests. Suggested review order:

1. specs/002-mock-call-real-crm/spec.md — the 35 FRs + 15 SCs are the contract.
2. specs/002-mock-call-real-crm/contracts/cli-slice2.md — Run Report JSON schema + state-machine spec + exit-status mapping.
3. src/opencloser/crm/dataverse/ — the SOLE location of Dataverse vendor detail (SC-010 enforced by tests/contract/test_boundary_isolation.py).
4. src/opencloser/slice2/runner.py + slice2/resume.py — Slice 2 CLI-level coordination; the Slice 1 orchestrator is reused unchanged (FR-014).
5. tests/contract/test_dataverse_adapter_contract.py (T017) — proves the adapter satisfies the unchanged Slice 1 crm-writeback.md contract for all 11 dispositions.
6. tests/integration/test_us*_*.py — end-to-end per user story against the in-process Dataverse fake.
7. specs/002-mock-call-real-crm/checklists/reverification.md — 72-item audit record; every gap was either closed inline or tracked as a numbered task.

Test plan

• CI: ruff + pytest green (uv run ruff check src/ tests/ && uv run pytest)
• Local: opencloser run-crm --queue-item-id <test-guid> produces planned writeback (dry-run default) against a verified mapping
• Local: opencloser run-crm --write --queue-item-id <test-guid> writes one Phone Call activity + one queue PATCH + one Task to a dedicated test record in Dynamics
• Manual: walk quickstart.md end-to-end (§3 discover → §5 dry-run → §6 write-enabled → §7 conflict/resume recovery → §8 cleanup)
• Verify GH issue Transport: validate fixture at place_call time to avoid partial session state #2 closed (done via T043 on 2026-05-24)

Deferred follow-up (out of scope for this PR)

• T052 — Persist planned-writeback.json sidecar BEFORE emit_* attempts, so a natural transient-exhaust failure leaves a replay-ready artifact. Documented in contracts/cli-slice2.md §Behavior — resume and pinned by test_us4_natural_transient_exhaust_yields_resume_needed.

🤖 Generated with Claude Code