proffesor-for-testing__agentic-qe/docs/releases/v3.7.3.md

v3.7.3 Release Notes

Release Date: 2026-02-27

Highlights

This release completes the Phase 0 + Phase 1 CI/CD integration milestone, making AQE CLI fully CI-native. Any CI/CD pipeline (GitHub Actions, GitLab CI, Jenkins, CircleCI, Buildkite) can now consume AQE outputs natively — no special SDK, no MCP requirement, no platform lock-in.

The strategic decision: CLI-first, format-rich, platform-agnostic. aqe security --format sarif runs in ANY pipeline and brings 150K learned patterns to bear.

Phase 0: CLI/MCP Verification — 17 Bug Fixes

Ran 87 test cases across all 26 CLI commands and 59 MCP tools. Found and fixed 17 issues (5 critical, 6 high, 6 medium):

Critical Fixes

• CRIT-1: CLI process hangs after command completion — switched to parseAsync() + cleanupAndExit()
• CRIT-2: Auto-init messages pollute stdout in --format json mode — routed to stderr
• CRIT-3: aqe quality calls wrong domain API method (evaluate → evaluateGate) — fixed with correct method signature and default metrics/thresholds
• CRIT-4: aqe security --compliance calls wrong method (checkCompliance → runComplianceCheck) — fixed with Promise.all for multiple frameworks
• CRIT-5: OpenRouter/Ollama providers crash on undefined defaultModel — added null coalescing

High-Priority Fixes

• HIGH-1: aqe <nonexistent> hangs instead of showing help
• HIGH-2: aqe status/aqe health auto-initializes in non-project dirs — added ensureInitializedStrict() checking for .agentic-qe/ dir
• HIGH-3: MCP aqe_health reports hardcoded version '3.7.0' — changed to build-time __CLI_VERSION__
• HIGH-4: team_scale ignores requested size due to clamping — maxSize = Math.max(getMaxTeamSize, targetSize, maxActiveTeams)
• HIGH-5: test_execute_parallel throws on failed tests instead of returning result — returns failed result object
• HIGH-6: infra_healing_status shows "not initialized" — added inline default playbook YAML fallback

Medium-Priority Fixes

• MED-1: Test plan references non-existent CLI flags
• MED-2: Coverage percentages show too many decimals — added .toFixed(1) formatting
• MED-3: aqe quality without --gate submits async task instead of inline eval — always uses inline mode now
• MED-4: Ghost coverage analyzer produces false positive security/integration gaps on math files — added getApplicableCategories() file-context filtering
• MED-5: model_route always returns tier 0 with complexity 0 — added minimum complexity floor, expanded keywords
• MED-6: MCP coverage_analyze_sublinear returns 0% vs CLI ~80% — added buildHeuristicCoverage() fallback

Phase 1: CI-Native CLI — All 5 Items Complete

1.1 --format flag on all output-producing commands ✅

All 4 core commands support format-aware output:

• aqe test execute --format json|junit|markdown
• aqe coverage --format json|markdown
• aqe security --format json|sarif|markdown
• aqe quality --format json|markdown

1.2 --output <path> for artifact generation ✅

All 4 core commands support --output <path> via shared writeOutput() utility. When used, writes full report to file and prints summary to stdout.

1.3 vitest JUnit XML generation ✅

Added JUnit reporter to vitest.config.ts. CI workflows now get junit.xml automatically.

1.4 Synchronous quality gate with direct pass/fail ✅

aqe quality now runs inline (no async task submission + polling). Returns structured JSON with criteria breakdown, per-criterion pass/fail, overall verdict, and exit code.

1.5 SARIF v2.1.0 output for aqe security ✅

toSARIF() fully implemented in ci-output.ts. Produces schema-compliant SARIF for GitHub Code Scanning / GHAS integration.

1.6 Exit codes: 0 / 1 / 2 convention ✅

Exit Code	Meaning	CI Usage
0	Success	Pipeline continues
1	Failure	Pipeline fails
2	Warning	Pipeline can choose to continue or fail

Per-command logic:

Command	Exit 1	Exit 2
security	Critical/high vulnerabilities	Medium-only vulnerabilities
quality	Quality gate failed	Score within 5% of perfect
coverage	Below threshold	Within 5% above threshold
test	Any test failures	>20% tests skipped

Bug fix: aqe coverage previously always exited 0 regardless of threshold — now correctly exits 1 when below threshold.

Additional Fixes

• concept_edges O(n^2) bloat — Prevented quadratic edge growth in knowledge graph and isolated test DB
• Crypto-secure randomness — Replaced Math.random() with crypto.randomUUID() in CI-sensitive paths
• OpenCode skill tier validation — Fixed 9 skill configs with invalid tier1-fast → tier1-any
• Score calculator tests — Updated test assertions for MED-5 minimum complexity floor

CI Pipeline Example

After this release, a complete AQE quality gate in GitHub Actions looks like:

- name: Security scan
  run: aqe security --sast --format sarif --output security.sarif

- name: Upload SARIF
  uses: github/codeql-action/upload-sarif@v3
  with:
    sarif_file: security.sarif

- name: Coverage check
  run: aqe coverage --threshold 80 --format json --output coverage.json

- name: Quality gate
  run: aqe quality --format json --output quality.json

- name: Test execution
  run: aqe test execute --format junit --output results.xml

Exit codes 0/1/2 let pipelines distinguish pass/fail/warn without parsing output.