AgentSkillsCN

nasa-safe-code-rater

依据NASA安全编码指南，对C/C++代码进行评分，采用10条归一化规则，生成0–100的安全评分，并以英文报告的形式呈现分析结果、排除项以及改进建议。当用户调用$nasa-safe-code-rater，或要求对某个代码仓库/当前文件/差异进行NASA风格的C语言安全编码合规性审计时，可使用此功能。

SKILL.md
--- frontmatter
name: nasa-safe-code-rater
description: Score C/C++ code against NASA safe coding guidance using 10 normalized rules, produce a 0-100 safety score, and return an English report with findings, exclusions, and remediation. Use when the user invokes $nasa-safe-code-rater or asks to audit a repository/current file/diff for NASA-style safe C coding compliance.

NASA Safe Code Rater

Use this skill to evaluate C/C++ code with a NASA-inspired 10-rule model and generate a scored report.

Inputs

Accept optional user arguments:

  • •target=repo|file|diff (default repo)
  • •path=<file-or-dir> (required for target=file)
  • •include_cpp=true|false (default true)

If arguments are absent, run repository-wide analysis.

Required Workflow

  1. •Resolve scope from user input.
  2. •Analyze only C/C++ files by default (.c, .h, .cpp, .hpp).
  3. •List non-C/C++ files as excluded items.
  4. •Evaluate all 10 rules in references/nasa_rules.md.
  5. •Produce an English markdown report containing:
  • •Executive summary
  • •Overall score and risk class (Good, Warning, Critical)
  • •Rule-by-rule findings table
  • •Top 5 remediation actions
  • •Insufficient evidence section

Execution Commands

Use these scripts from the skill directory:

bash
python3 scripts/analyze_repo.py --root <repo-path> --mode repo --json-out /tmp/nasa_analysis.json
python3 scripts/score_report.py --analysis /tmp/nasa_analysis.json --format markdown

For single file mode:

bash
python3 scripts/analyze_repo.py --root <repo-path> --mode file --path <file-path> --json-out /tmp/nasa_analysis.json

For diff mode:

bash
python3 scripts/analyze_repo.py --root <repo-path> --mode diff --json-out /tmp/nasa_analysis.json

To refresh normalized rules from a source PDF:

bash
python3 scripts/extract_nasa_rules.py --pdf <path-to-pdf> --out references/nasa_rules.md

Quality Gates

Require for critical findings:

  • •At least one concrete evidence line per failed rule.
  • •Rule IDs traceable to references/nasa_rules.md.

If no C/C++ files are in scope, return not applicable with reason and next step.