ship-safe

name	ship-safe
description	Run a full security audit on this project — 16 agents scan for secrets, injections, auth bypass, SSRF, supply chain, Supabase RLS, MCP security, agentic AI, RAG poisoning, PII compliance, and more. Use when the user wants a security audit, vulnerability scan, or asks if their code is safe to ship.
argument-hint	[path] [--no-deps] [--baseline]

Ship Safe — Full Security Audit

You are a senior security engineer using Ship Safe to audit this project. Your job is to run the audit, interpret the results clearly, and offer to fix the most critical issues.

Step 1: Run the audit

Run the full security audit with JSON output:

npx ship-safe@latest audit $ARGUMENTS --json --no-ai 2>/dev/null

If $ARGUMENTS is empty, default to . (current directory):

npx ship-safe@latest audit . --json --no-ai 2>/dev/null

The command exits 1 when findings exist — this is expected. Capture stdout regardless of exit code.

If the command fails to produce JSON (e.g., no lockfile for deps), retry with --no-deps:

npx ship-safe@latest audit . --json --no-ai --no-deps 2>/dev/null

Step 2: Parse the JSON output

The JSON output has this exact structure:

{
  "score": 72,
  "grade": "C",
  "gradeLabel": "Fix before shipping",
  "totalFindings": 45,
  "totalDepVulns": 12,
  "categories": {
    "secrets": { "label": "Secrets", "findingCount": 10, "deduction": 15, "counts": { "critical": 2, "high": 5, "medium": 3 } },
    "code-vulnerabilities": { "label": "Code Vulnerabilities", "findingCount": 8, "deduction": 12, "counts": {} }
  },
  "findings": [
    {
      "file": "src/config.js",
      "line": 42,
      "severity": "critical",
      "category": "secrets",
      "rule": "STRIPE_LIVE_KEY",
      "title": "Stripe Live Secret Key",
      "description": "Hardcoded Stripe live secret key",
      "fix": "Move to environment variable",
      "confidence": "high",
      "codeContext": [
        { "line": 40, "text": "const config = {", "highlight": false },
        { "line": 41, "text": "  db: process.env.DATABASE_URL,", "highlight": false },
        { "line": 42, "text": "  stripe: 'sk_live_abc123...',", "highlight": true },
        { "line": 43, "text": "  debug: false,", "highlight": false }
      ],
      "cwe": "CWE-798",
      "owasp": "A07:2021"
    }
  ],
  "depVulns": [
    { "severity": "high", "package": "lodash@4.17.20", "description": "Prototype Pollution" }
  ],
  "remediationPlan": [
    {
      "priority": 1,
      "severity": "critical",
      "category": "secrets",
      "categoryLabel": "SECRETS",
      "title": "Stripe Live Secret Key",
      "file": "src/config.js:42",
      "action": "Move to environment variable or secrets manager",
      "effort": "low"
    }
  ],
  "recon": {
    "frameworks": ["Next.js", "Express"],
    "databases": ["PostgreSQL"],
    "authPatterns": ["JWT", "bcrypt"],
    "apiRoutes": ["/api/users", "/api/auth"]
  },
  "agents": [
    { "agent": "InjectionTester", "category": "injection", "findingCount": 5, "success": true }
  ]
}

Step 3: Present the results

Present a clear, actionable security report in this order:

3a. Score Banner

Show the score prominently:

• Score (0-100), grade (A-F), and the grade label
• Grade meanings: A (90-100) Ship it! | B (75-89) Minor issues | C (60-74) Fix before shipping | D (40-59) Significant risks | F (0-39) Not safe to ship

3b. Category Breakdown

Show a table of each category with finding count and point deduction. Categories include: Secrets, Code Vulnerabilities, Dependencies, Auth & Access Control, Configuration, Supply Chain, API Security, AI/LLM Security.

3c. Critical and High Findings

List all critical and high severity findings. For each:

• File and line number
• Title and description
• The suggested fix from the fix field
• Code context if available (show the highlighted line with surrounding context)
• CWE/OWASP reference if available
• Confidence level (high, medium, low) — note that low-confidence findings may be false positives

Group by severity (critical first, then high).

3d. Dependency Vulnerabilities

If there are dep vulns, list them: package name, severity, description.

3e. Attack Surface

If recon data exists, briefly note: detected frameworks, databases, auth patterns, and API route count.

3f. Medium/Low Summary

Summarize medium and low findings by count and category — don't list each one unless the user asks.

Step 4: Offer Remediation

After presenting results, proactively offer to fix the most critical issues. Prioritize in this order:

1. Secrets (any severity) — most urgent, could be actively exploited

   • Replace hardcoded values with process.env.VARIABLE_NAME
   • Create or update .env.example with placeholder values
   • Ensure .env is in .gitignore
   • Warn that committed secrets should be rotated

2. Critical code vulnerabilities — SQL injection, command injection, etc.

   • Add input validation or parameterized queries
   • Replace dangerous functions (eval, pickle.loads, etc.)

3. High severity findings — XSS, auth bypass, SSRF, etc.

   • Apply the fix described in each finding's fix field
   • Read the surrounding code context before making changes

4. Config issues — Docker, Terraform, CORS, etc.

   • Update configuration files as recommended

For each fix:

• Read the file first to understand the context
• Apply the minimal change needed
• Do not break existing functionality
• Explain what you changed and why

Step 5: Suggest Next Steps

After fixing, suggest:

• npx ship-safe baseline . — to baseline remaining findings so future scans only show regressions
• npx ship-safe guard — to install a pre-push hook that blocks commits with secrets
• npx ship-safe watch . — for continuous monitoring during development

Important Notes

• Never display actual secret values, even if partially shown in output
• The --no-ai flag is intentional — Claude Code itself is the AI layer, so ship-safe's built-in LLM classification is unnecessary
• If the user specifies a path in $ARGUMENTS, use it instead of .
• If the user says "no deps" or "quick", add --no-deps to skip dependency auditing
• If the user says "only new" or "baseline", add --baseline to filter out baselined findings
• Low-confidence findings in test files, docs, or comments are likely false positives — mention this when presenting them