It seems I'm being blocked from writing files. Let me present the plan directly instead.

---

Add an integration test suite that goes beyond the existing 17 mock-based test suites by running the real pipeline with enhanced mock binaries that validate stage ordering, state transitions, and artifact generation — plus a CI workflow with budget enforcement and structured reporting.

Two tiers:

1. Smoke tests (no Claude, no GitHub) — validate pipeline machinery with deterministic mocks
2. Live integration tests (requires CLAUDE_API_KEY + GITHUB_TOKEN) — run a minimal pipeline against real Claude, budget-capped at 1ドル.00

---

File	Action	Purpose
scripts/sw-integration-test.sh	Create	Integration test suite (~500 lines)
templates/pipelines/integration-test.json	Create	Minimal 3-stage pipeline template
.github/workflows/integration-test.yml	Create	CI workflow (smoke + live jobs)
.github/workflows/test.yml	Modify	Add smoke test step after unit tests
package.json	Modify	Add test:smoke and test:integration scripts

---

Minimal template: intake → build → test only, all auto-gated, model sonnet, max 3 iterations.

Same harness pattern as sw-pipeline-test.sh: set -euo pipefail, ERR trap, PASS/FAIL counters, temp dir isolation, mock binaries on PATH, assertion functions.

Tier 1 — Smoke Tests (10 tests, always run):

1. test_smoke_full_stage_ordering — Run fast template E2E, parse pipeline-state.md, verify stage_progress order, status: success
2. test_smoke_state_transitions — Verify state transitions pending → running → complete per stage, current_stage updates correctly
3. test_smoke_artifact_integrity — Verify all expected artifacts exist: intake.json (valid JSON), plan.md, test-results.log, events.jsonl entries
4. test_smoke_no_crashes_fast_template — Fast template exits 0, no ERROR lines
5. test_smoke_no_crashes_standard_template — Standard template exits 0
6. test_smoke_no_crashes_autonomous_template — Autonomous template exits 0
7. test_smoke_budget_enforcement — Mock cost at 1ドル.01, verify pipeline fails; mock at 0ドル.99, verify it passes
8. test_smoke_resume_preserves_artifacts — Fail at test stage, resume, verify prior artifacts preserved
9. test_smoke_dry_run_no_side_effects — --dry-run creates no artifacts, no branches, no events
10. test_smoke_stage_timing_recorded — Each completed stage has duration in state file log

Tier 2 — Live Tests (2 tests, require secrets):

11. test_live_trivial_readme_change — Real Claude makes a trivial change, verify diff exists, clean git status, cost < 1ドル.00
12. test_live_pr_creation — Verify PR created with valid URL, cleanup after

Mode selection: --live flag enables Tier 2, --filter <name> for single test.

• smoke job: Matrix (macOS + Ubuntu), runs on every PR and push to main
• live job: Runs after smoke passes, only on push to main or manual trigger, uses integration-test environment for secret access, 15-minute timeout, uploads artifacts

Add "test:smoke" and "test:integration" scripts.

Add - name: Run integration smoke tests step after existing unit test steps.

---

• Task 1: Create templates/pipelines/integration-test.json
• Task 2: Create scripts/sw-integration-test.sh scaffold (harness, mocks, assertions)
• Task 3: Implement smoke tests 1-5 (ordering, transitions, artifacts, crash tests)
• Task 4: Implement smoke tests 6-10 (budget, resume, dry-run, timing)
• Task 5: Implement live tests (README change, PR creation)
• Task 6: Create .github/workflows/integration-test.yml
• Task 7: Update package.json with new scripts
• Task 8: Update .github/workflows/test.yml with smoke step
• Task 9: Add CI summary reporting ($GITHUB_STEP_SUMMARY markdown table)
• Task 10: Run smoke tests locally and fix failures
• Task 11: Verify CLAUDE.md AUTO sections update

---

1. Local: bash scripts/sw-integration-test.sh — all 10 smoke tests pass
2. Filter: --filter test_smoke_budget_enforcement for debugging individual tests
3. CI: Push branch → verify workflow triggers and summary reports
4. Live: Manual dispatch with run_live: true or merge to main
5. Cross-platform: macOS + Ubuntu matrix

---

• npm run test:smoke passes (exit 0)
• npm run test:integration runs full suite when secrets available
• CI runs smoke tests on every PR to main
• CI runs live tests on push to main (regression)
• Budget cap: live tests abort if cost > 1ドル.00
• CI summary: pass/fail per test in $GITHUB_STEP_SUMMARY
• Existing 17 test suites still pass
• Follows conventions: set -euo pipefail, bash 3.2, shipwright color theme