跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

meddev-code-coverage

Measure and enforce coverage appropriate to safety class: statement/branch and MC/DC where applicable, with documented exclusions. Use when working on code coverage in medical device software.

AminAlam · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add AminAlam/meddev-agent-skills --skill meddev-code-coverage

技能说明


name: meddev-code-coverage description: 'Measure and enforce coverage appropriate to safety class: statement/branch and MC/DC where applicable, with documented exclusions. Use when working on code coverage in medical device software.' license: MIT metadata: skill_id: TEST-COVERAGE version: 1.1.0 last_updated: '2026-05-21' applies_to: Class A, Class B, Class C jurisdiction: Global prerequisites: TEST-UNIT, TEST-INTEGRATION category: testing

Code Coverage Analysis

Purpose

Measure and enforce coverage appropriate to safety class: statement/branch and MC/DC where applicable, with documented exclusions.

When to Apply

  • Unit and integration testing phases; before releases.
  • After significant code changes affecting control logic.

Requirements (testable)

  1. Metrics: Collect statement and branch coverage; for Class C control logic, pursue MC/DC where feasible (informative practice; not a separate normative requirement in IEC 62304:2006). Rationale: evidence of test thoroughness.
  2. Targets by Class: Define minimums (e.g., Class A: stmt≥80%, branch≥70; Class B: stmt≥90%, branch≥80; Class C: stmt≥95%, branch≥90, MC/DC for safety logic). Rationale: risk-based rigor.
  3. Exclusions: Document and justify any excluded code (e.g., defensive assertions, unreachable hardware stubs); peer review exclusions. Rationale: transparency.
  4. Tooling: Use consistent coverage tools; ensure build flags don’t alter behavior materially; keep artifacts. Rationale: reproducibility.
  5. Gating: Block merges/releases when coverage regresses beyond thresholds without approved justification. Rationale: maintain bar.

Recommended Practices

  • Separate coverage runs for unit vs integration if needed; merge reports.
  • Include tests for error paths to improve branch coverage.
  • Periodically review high-risk modules for MC/DC attempts even if tooling is limited.

Patterns

Coverage thresholds (example policy):

Class B min: stmt 90%, branch 80%; MC/DC encouraged for control logic

Exclude rationale entry:

# REQ-COV-EXCL-01
file: src/assert.c
region: fatal_halt()
reason: Safety halt path; not reachable in normal test. Verified by inspection.
approved_by: safety_reviewer

Anti-Patterns (risks)

  • Counting coverage with tests that do not assert outcomes -> risk: false confidence.
  • Large unreviewed exclusion lists -> risk: untested critical logic.
  • Using coverage flags that change optimization/behavior in unsafe ways -> risk: mismatch to production.

Verification Checklist

  • Coverage collected (stmt/branch; MC/DC where applicable) with defined thresholds by class.
  • Exclusions documented, reviewed, and minimal.
  • Coverage reports stored as artifacts; reproducible runs.
  • Coverage gating enforced; regressions blocked or justified/approved.
  • Error/edge paths tested to support branch/MC/DC.

Traceability

  • REQ-COV-### linked to coverage policy and reports per release; map safety-critical modules to MC/DC efforts.

References

  • IEC 62304:2006+A1:2015, 5.5 (unit verification); Annex A Table A.1 (informative class tailoring).
  • DO-178C MC/DC concepts (informative; not required by IEC 62304:2006).
  • TEST-UNIT.
  • Tool docs (gcov/lcov, llvm-cov, commercial MC/DC tools).

Changelog

  • 1.1.0 (2026-05-21): Clarified MC/DC as informative; linked 5.5 and class tailoring.
  • 1.0.0 (2026-01-04): Initial coverage skill with class-based thresholds and exclusion governance.

Audit History

  • 2026-01-04: Audit performed. Verified:
    • MC/DC concepts correctly referenced from DO-178C as informative
    • Coverage thresholds by safety class are reasonable guidelines
    • gcov/lcov and llvm-cov tool references are accurate