D
Datarim
EN RU
Command Pipeline

/dr-qa

Multi-layer quality verification across PRD, design, plan, and code

Overview

/dr-qa runs a four-layer quality verification that checks whether the implementation actually matches what was planned, designed, and required. It resolves the target task first, then each layer runs conditionally — only if the relevant artifacts exist.

Usage

/dr-qa

The Four Layers

  1. Layer 1: PRD Alignment — compares stated requirements against implementation. Flags missing features and scope creep. Only runs if datarim/prd/*.md exists.
  2. Layer 2: Design Conformance — verifies that architectural decisions (ADRs) are respected in code and chosen patterns are actually used. Only runs if datarim/creative/*.md exists.
  3. Layer 3: Plan Completeness — checks that each planned implementation step was completed, flags skipped or unplanned steps. Only runs if tasks.md has implementation steps.
  4. Layer 4: Code Quality — always runs. Includes five sub-checks:
    • 4a. Tests — runs the project test suite and reports pass/fail/skip counts
    • 4b. Security — scans for hardcoded secrets, exposed endpoints, missing input validation, dependency vulnerabilities
    • 4c. Anti-Patterns — methods over 50 lines, 7+ parameters, duplicated code, missing error handling, stray console.log statements
    • 4d. Live Smoke-Test Gate — if changed code uses $queryRaw, raw(), sequelize.query(), or any path bypassing the ORM type-checker, a live smoke test against the actual target datasource is mandatory. Mocked/unit tests do not satisfy this gate. In multi-datasource projects, verifies the correct client is injected. The exact smoke-test command, datasource hit, and result are recorded in the QA report. No smoke test = Layer 4 verdict is FAIL.
    • 4e. Definition of Done — reads DoD from tasks.md or prd/*.md and checks each criterion

CLI Agent Exit Code Pattern

Many CLI agents return exit code 0 even when the JSON output contains is_error: true. When testing integrations with CLI agents, check is_error/subtype in JSON — do not rely on exit codes for pass/fail determination.

Arguments

None. Operates on the currently active task (resolved via Task Resolution).

Output

Creates: datarim/qa/qa-report-{task-id}.md with per-layer verdicts.

Verdicts

  • PASS — all checks satisfied, no issues
  • PASS_WITH_NOTES — all checks satisfied, minor observations that do not block
  • FAIL — one or more checks failed, must be addressed

Overall: ALL_PASS (proceed), CONDITIONAL_PASS (proceed with notes), or BLOCKED (return to /dr-do with fix list).

Example Session

> /dr-qa

Task Resolution: AUTH-0001 (Level 3)

Layer 1: PRD Alignment — PASS
  8/8 requirements implemented
  No scope creep detected

Layer 2: Design Conformance — PASS_WITH_NOTES
  ADR-001 (WebSocket): Followed
  ADR-002 (Redis cache): Followed
  Note: Connection pooling added (not in design, but improvement)

Layer 3: Plan Completeness — PASS
  8/8 steps completed
  No skipped steps

Layer 4: Code Quality — PASS
  4a. Tests: 14 passed, 0 failed
  4b. Security issues: 0
  4c. Anti-patterns: 0
  4d. Smoke test: SELECT 1 on auth DB — OK (1 row)
  4e. DoD: 5/5 criteria met

Overall: ALL_PASS
Report saved: datarim/qa/qa-report-AUTH-0001.md

Next step: /dr-compliance or /dr-archive

Related Commands

← All Features