Skip to content

SCOPE.md

Latest commit

 

History

History
80 lines (68 loc) · 3.77 KB

SCOPE.md

File metadata and controls

80 lines (68 loc) · 3.77 KB
Raw

Scope: What TestAtlas Will Refuse To Do

TestAtlas is opinionated about what it is not. This document codifies the scope-rejection list referenced in CONTRIBUTING.md and enforces requirement GOV-04. New feature proposals that conflict with anything below will be closed.

This list combines:

  • The original PRD out-of-scope table
  • Research-derived anti-features
  • PRD §42 non-goals

Hard Rejections (PROJECT.md Out of Scope)

TestAtlas will NOT:

  1. Replace existing test runners (Jest, Playwright, Pytest, etc.). TestAtlas guides agents to use them; it does not reimplement them.
  2. Test production by default. allowProductionTesting: false is the safe default; opt-in only.
  3. Cause real email/SMS/payment side effects. Sandbox-only by safety constraint.
  4. Store secrets in _testatlas/. The redaction pipeline strips secrets/PII from evidence; storing secrets is forbidden.
  5. Issue formal compliance certification. TestAtlas surfaces a11y/security risks; it does not certify SOC2, HIPAA, GDPR, etc.
  6. Make scripts mandatory. JS scripts are optional accelerators; the suite must work agent-only (PRD §22).
  7. Require browser automation as a hard dependency. Chrome DevTools MCP is first-class but graceful fallback when unavailable.
  8. Auto-apply updates without user prompt. Risky for in-flight runs; deferred to v2 with guardrails.
  9. Support private / PAT-gated installs in v1. Public GitHub only.
  10. Auto-fix issues. TestAtlas surfaces issues; remediation is a separate agent's responsibility.
  11. Optimize for a single platform. Tool-agnostic principle requires breadth across all 7 adapter families.
  12. Provide a test-as-code DSL. Markdown templates + JSON schemas are the contract.
  13. Overwrite human content without generated-section markers. Generated-section discipline is enforced.

Anti-Features (Research-Derived Rejections)

TestAtlas will NOT:

  1. Ship a TypeScript build step in v1. Plain ESM JavaScript per PRD §22 direct-invocation contract; revisit with V2-01 once Node 24 native strip-types is stable.
  2. Use CommonJS. ESM-only ("type": "module").
  3. Bundle / transpile suite scripts. node script.js runs raw.
  4. Depend on chalk, fs-extra, update-notifier, node-fetch, axios, inquirer, ajv@6, commander<13, husky. Native APIs and smaller alternatives are mandated by research/STACK.md.
  5. Use a bash-only installer. install.sh is POSIX /bin/sh, shellcheck-clean.
  6. Allow hand-edited derived adapters. Generated by assemble-adapter.js from the canonical Claude Code spec; CI rejects hand-edits.
  7. Allow bootstrap.md to exceed ~3000 words. Token-budget CI gate enforces.
  8. Allow any single command file to exceed ~1800 words. Same gate.
  9. Allow findings without evidence. "No evidence, no finding" is a hard rule in bootstrap (first 500 tokens) and enforced by validate-workspace.
  10. Fabricate output when a required capability is unavailable. Capability- aware degradation rule emits explicit "tool unavailable" notes; affected findings marked confidence: needs-validation.
  11. Run a real-time test dashboard or web UI in v1. Deferred to V2-03.
  12. Sync workspace to cloud in v1. Deferred to V2-02.
  13. Provide opt-in telemetry in v1. Deferred to V2-06.
  14. Localize for non-English teams in v1. Deferred to V2-05.
  15. Provide a built-in test-runner shim in v1. Deferred to V2-08.

How to Propose New Scope

If you believe a rejected item should be reconsidered:

  1. Open a GitHub Discussion (not an Issue) with subject [scope] <topic>.
  2. Cite the rejection number above and explain what changed.
  3. Maintainer consensus required to amend this document.

See Also

  • docs/THREAT_MODEL.md — Security boundaries that complement scope boundaries