Dragan Spiridonov
98a6b19c88
Ports ruflo's 'smoke.sh as a plugin contract' discipline. AQE already ships
the marketplace.json + a lean auto-discovery plugin (plugins/agentic-qe-fleet:
11 agents, 9 commands, 9 skills, lean plugin.json), but had no machine-checkable
contract — so silent drift could ship a broken bundle.
plugins/agentic-qe-fleet/scripts/smoke.sh (npm run plugin:smoke) asserts:
- plugin.json valid JSON, name == agentic-qe-fleet, has a version;
- plugin.json does NOT declare skills/commands/agents arrays (Claude Code
auto-discovers them — declaring them is an error, per ruflo);
- marketplace.json indexes the plugin at ./plugins/agentic-qe-fleet;
- exact counts (11 agents / 9 commands / 9 skills), each with name+description
frontmatter / SKILL.md (bump the constants deliberately on change);
- README present; top-level plugin.json registers the agentic-qe MCP server.
Exit 0 = contract holds, 1 = any violation.
Verified: passes 10/10 checks (exit 0), and correctly FAILS (exit 1) when a
skill is removed ('skills: 8 (expected 9)').
Scope note: AQE ships ONE bundle today. Splitting the 93 QE skills into multiple
lean subset bundles (aqe-security, aqe-browser, ...) is a larger restructure of
shipped artifacts — deferred as a follow-up (needs a distribution decision); the
skills-manifest.json already has the category data to drive it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
agentic-qe-fleet
PACT-based agentic quality engineering fleet for Claude Code — a slim starter bundle of the AQE platform with 11 specialized QE agents, 9 core skills, 9 slash commands, and the agentic-qe MCP server.
What's bundled
Agents (11)
Routed by cognitive load — heavy reasoning agents on Opus, focused execution on Sonnet.
| Agent | Model | Purpose |
|---|---|---|
qe-test-architect |
opus | AI-powered test generation with sublinear optimization |
qe-fleet-commander |
opus | Fleet lifecycle and workload distribution |
qe-security-scanner |
opus | SAST/DAST/dependency/secrets scanning |
qe-chaos-engineer |
opus | Controlled fault injection and resilience testing |
qe-regression-analyzer |
opus | Intelligent test selection and change-impact scoring |
qe-requirements-validator |
opus | Testability analysis and BDD scenario generation |
qe-coverage-specialist |
sonnet | O(log n) sublinear coverage analysis with risk-weighted gap detection |
qe-flaky-hunter |
sonnet | Flaky test detection and auto-stabilization |
qe-performance-tester |
sonnet | Load, stress, endurance, regression detection |
qe-quality-gate |
sonnet | Quality gate enforcement with policy validation |
qe-tdd-specialist |
sonnet | Red-Green-Refactor (London + Chicago schools) |
Skills (9)
Each skill ships with a scoped allowed-tools list and a trust tier. Tier 3 = full eval infrastructure (eval YAML + JSON schema + validator); tier 2 = tested without eval. The bundle excludes tier-1 (untested) skills per AQE trust-tier policy.
| Skill | Trust tier | Notes |
|---|---|---|
qe-test-generation |
3 | Includes mcp__agentic-qe__test_generate_enhanced |
qe-coverage-analysis |
3 | Includes coverage_analyze_sublinear, qe_coverage_gaps |
qe-test-execution |
3 | Includes test_execute_parallel |
qe-chaos-resilience |
3 | Includes chaos_test |
qe-quality-assessment |
3 | Includes quality_assess |
chaos-engineering-resilience |
3 | Methodology guide |
mutation-testing |
3 | Stryker integration |
risk-based-testing |
3 | Read-only analysis |
tdd-london-chicago |
2 | TDD school comparison |
Commands (9)
/aqe-analyze, /aqe-benchmark, /aqe-chaos, /aqe-costs, /aqe-execute, /aqe-fleet-status, /aqe-generate, /aqe-optimize, /aqe-report
MCP server
The plugin auto-registers the agentic-qe MCP server (via npx -y agentic-qe@latest mcp) when loaded — no separate claude mcp add needed.
Test locally
claude --plugin-dir ./plugins/agentic-qe-fleet
Then invoke an agent or skill, for example:
/aqe-fleet-status
Install from a marketplace
Once published, users install with:
/plugin marketplace add <org>/<repo>
/plugin install agentic-qe-fleet