Add trust command for package trust verification

[thebenignhacker]

Summary

• Add hackmyagent trust <package> for single package trust lookup
• Add hackmyagent trust --audit <file> to scan package.json/requirements.txt dependencies
• Add hackmyagent trust --batch pkg1 pkg2 for multi-package queries
• Queries OpenA2A Registry trust API (same endpoints as ai-trust CLI)
• Build passes, 908 existing tests pass

Test plan

• npm run build clean
• All 908 existing tests pass
• Smoke tested all 3 modes against live registry

[github-actions]

Claude Code Review

Security Review: PR #44 - Add trust command for package trust verification

VERDICT: APPROVE

SUMMARY:

This PR adds a trust command for querying the OpenA2A Registry API to check trust scores for AI packages. The implementation includes three modes: single package lookup, batch lookup, and dependency file auditing. After thorough verification of all potential security concerns, I found no exploitable vulnerabilities. All user inputs are properly validated, external API calls are safe, file operations are protected, and the code follows secure patterns throughout.

FINDINGS:

None. All initially identified concerns were mitigated upon verification:

Verified Safe Patterns:

1. URL Construction (lines 4649, 4669) - VERIFIED SAFE

   • registryUrl is sanitized via .replace(/\/+$/, '') at line 4908 before use
   • User-controlled name parameter is passed through URLSearchParams constructor which automatically URL-encodes values
   • No string concatenation or template injection possible

2. Arbitrary File Read (line 4830) - VERIFIED SAFE

   • parseDepsFile() receives filePath from Commander option --audit <file>
   • Commander treats option values as raw strings (no shell expansion)
   • User explicitly provides the path they want to read (intended functionality)
   • No path traversal validation needed - user chooses which file to audit
   • Only supports package.json/requirements.txt (line 4869 throws on others)

3. JSON Parsing (lines 4660, 4679, 4836) - VERIFIED SAFE

   • Line 4660: Parses response from Registry API (external but HTTPS, not arbitrary user input)
   • Line 4679: Same as above (batch endpoint)
   • Line 4836: Parses user's own package.json file (user chose to audit their own file)
   • Standard JSON.parse() errors are caught by outer try/catch (line 4915)

4. ReDoS in Requirements Parsing (line 4857) - VERIFIED SAFE

   • Regex: /^([a-zA-Z0-9_-]+(?:\[[a-zA-Z0-9_,-]+\])?)/
   • Pattern is linear-time: single character class with optional bracketed section
   • No nested quantifiers or catastrophic backtracking paths
   • Bounded by line splits (one match per line)

5. Integer Parsing (line 4909) - VERIFIED SAFE

   • parseInt(opts.minTrust, 10) with base-10 specified
   • Validated: isNaN(minTrust) || minTrust < 0 || minTrust > 4 (lines 4910-4913)
   • Safe range enforcement before use

6. SSRF via Registry URL (lines 4649, 4669) - VERIFIED ACCEPTABLE

   • Default: https://api.oa2a.org (hardcoded constant line 4612)
   • User can override via --registry-url flag
   • Design Intent: Users should be able to point to alternate/dev registries
   • This is a CLI tool where users control their own environment
   • No automatic execution of untrusted URLs

All security boundaries are correctly enforced. The code properly handles user input, validates ranges, and safely interacts with external APIs.

---

Reviewed 3 files changed (15246 bytes)