Quality Stack
Scan a project's build configuration across JVM, Android, Node.js, and Python ecosystems, cross-reference against curated research documents, and assist with tool setup.
Pre-flight
- Run the orchestrator — it auto-detects ecosystems:
``bash python3 <skill-path>/scripts/scan_project.py <project-root> ``
- If
"error": "no_ecosystem_detected", checknearby_project_filesfor subproject paths. - Monorepo? — use
--recursiveor--ecosystemto force a specific scanner:
``bash python3 <skill-path>/scripts/scan_project.py --recursive <project-root> python3 <skill-path>/scripts/scan_project.py --ecosystem node <project-root> ``
- Legacy (JVM only) —
scan_tooling.pystill works as a backwards-compatible wrapper.
Two-Phase Workflow
Phase 1: Recommend
- Run the scanner on the project root (see Pre-flight above).
- Fetch research documents via WebFetch — only for detected ecosystems:
Android (when ecosystems contains "android"): `` https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/android-ecosystem-tooling.md https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/android-testing-ecosystem.md ``
JVM (when ecosystems contains "jvm"): `` https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/jvm-quality-tools-evaluation.md https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/kotlin-spring-boot-testing-ecosystem.md ``
Node.js (when ecosystems contains "node"): `` https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/node-quality-tools-evaluation.md ``
Python (when ecosystems contains "python"): `` https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/python-quality-tools-evaluation.md ``
Cross-cutting (always): `` https://raw.githubusercontent.com/joaquimscosta/arkhe-claude-plugins/main/docs/research/cross-cutting-devtools-evaluation.md ``
If WebFetch fails, warn the user and proceed using scanner results + LLM knowledge only.
- Cross-reference scanner output against research recommendations per ecosystem:
- Identify tools recommended but missing from the project
- Check
statusfield:disabledorconfig-onlytools need attention - Flag outdated or superseded tools
- Apply ecosystem-specific SKIP rules (see [WORKFLOW.md](WORKFLOW.md))
- Review
tool_configfor threshold values and settings
- Generate the recommendation report using the format in [WORKFLOW.md](WORKFLOW.md).
Phase 2: Setup
After presenting the report, present tools for selection using the multi-round protocol in [WORKFLOW.md](WORKFLOW.md). Group by ecosystem and priority tier. Include effort estimates. After user completes selection across all rounds:
- Read the relevant research doc section for setup instructions
- Apply [Setup Guards](WORKFLOW.md) — resolve versions, check compatibility
- For each selected tool, apply changes per ecosystem:
- JVM: Add Gradle/Maven plugin, test deps, config files
- Node.js:
pnpm add -D, tsconfig edits, config file creation - Python:
uv add --dev, pyproject.toml edits, config file creation - Cross-cutting: CI/CD workflow steps, Lefthook config, EditorConfig
- Verify each tool after configuration — run the tool's check command,
verify filter patterns against actual codebase paths, and check for config inheritance conflicts. See [WORKFLOW.md](WORKFLOW.md) Post-Setup Verification.
- Re-run the scanner to confirm all tools detected
Priority Classification
| Priority | Criteria | |----------|----------| | NOW | Essential missing tools, zero-dependency additions | | SOON | High-value additions requiring minor setup | | LATER | Nice-to-have with prerequisites | | SKIP | Not applicable (wrong ecosystem, incompatible version, deprecated) |
Ecosystem-aware rules — see [WORKFLOW.md](WORKFLOW.md) for full classification tables per ecosystem.
Key rules:
- Android Compose project: NOW Compose UI testing, SOON Roborazzi; SKIP Espresso
- Android KMP project: NOW commonTest setup, NOW Turbine; SOON Ktor MockEngine
- Android no lint config: NOW Android Lint baseline; SOON custom lint rules
- JVM Pure Kotlin: SKIP Error Prone, SpotBugs; JVM Pure Java: SKIP Detekt, ktlint, MockK
- JVM Spring Boot 4+: SKIP REST Assured, NOW MockMvcTester
- Node.js no linter: NOW ESLint; no formatter + no Biome: NOW Prettier
- Node.js TypeScript not strict: NOW enable strict
- Python no linter: NOW Ruff; no type checker + has type annotations: NOW mypy
Cross-Cutting Tools
| Tool | When to Recommend | Priority | |------|------------------|----------| | Lefthook | No git hook manager + has linters | SOON | | commitlint | No commit conventions + has team | LATER | | EditorConfig | Missing .editorconfig | NOW | | Renovate/Dependabot | No dependency automation | SOON | | Trivy/gitleaks | No security scanning | SOON |
Research Documents
Fetch via WebFetch at runtime — only for detected ecosystems:
- Android Ecosystem Tooling:
android-ecosystem-tooling.md - Android Testing Ecosystem:
android-testing-ecosystem.md - JVM Quality Tools:
jvm-quality-tools-evaluation.md - JVM Testing Ecosystem:
kotlin-spring-boot-testing-ecosystem.md - Node.js Quality Tools:
node-quality-tools-evaluation.md - Python Quality Tools:
python-quality-tools-evaluation.md - Cross-Cutting Tools:
cross-cutting-devtools-evaluation.md
Scanner Architecture
scripts/
scan_project.py # Orchestrator — auto-detects + merges
scan_jvm.py # JVM scanner (Gradle/Maven)
scan_android.py # Android scanner (AGP/Compose/KMP)
scan_node.py # Node.js/TypeScript scanner
scan_python.py # Python scanner
scan_cross_cutting.py # Cross-cutting tools (CI, hooks, security)
shared.py # Shared utilities
scan_tooling.py # Legacy wrapper → scan_jvm.py
References
- Workflow: See [WORKFLOW.md](WORKFLOW.md) for classification rules and report format
- Examples: See [EXAMPLES.md](EXAMPLES.md) for realistic audit scenarios
- Troubleshooting: See [TROUBLESHOOTING.md](TROUBLESHOOTING.md) for scanner issues