Appearance

Star

Security first artifact consumption ​

Scan and verify third party artifacts before they touch your codebase. Enforce security thresholds in CI to block risky artifacts from reaching production. Trust but verify.

The problem ​

Third-party AI configurations are powerful but risky. An artifact that instructs your AI tool could:

  • Contain prompt injection that overrides your safety rules
  • Include instructions to exfiltrate data through crafted outputs
  • Introduce subtle behavioral changes that are hard to notice
  • Ship malicious updates in a patch version that bypasses review

You want to use community artifacts, but you need to know what you're installing.

The solution ​

1. Scan before installing ​

Use grekt scan to analyze an artifact before adding it:

bash
grekt scan @community/code-assistant

The scanner uses AgentVerus (local, zero network) to check for prompt injection, data exfiltration, and other risk patterns. Artifacts on the public registry also get Snyk AI powered analysis for tool poisoning and toxic flows.

2. Review scan results ​

Artifacts receive a trust badge based on their AgentVerus score:

  • Certified (90+) - No high severity findings
  • Conditional (75-89) - Minor concerns
  • Suspicious (50-74) - Multiple findings
  • Rejected (<50) - Critical findings

Snyk adds risk labels like destructive, private_data, untrusted_content, and is_public_sink to flag specific behaviors.

3. Use trust signals ​

Look for trust badges when browsing the registry:

bash
grekt info @trusted-org/vetted-agent

Published artifacts carry their scan results, so you can evaluate risk before installing.

4. Enforce in CI ​

Individual scanning works for you, but teams have a different problem: someone adds an artifact without checking, pushes, and now malicious prompts are in your codebase.

Use --fail-on with HMAC signed trust to gate your pipeline while allowing reviewed exceptions. See Security gating for the full setup.

The short version:

bash
# Repo owner generates a trust key once
grekt trust --generate-key

# Repo owner signs reviewed artifacts
grekt trust @sketchy/tool

# CI verifies signatures and blocks unsigned risky artifacts
grekt scan --fail-on suspicious

5. Handle known exceptions ​

Sign reviewed artifacts with grekt trust. Only the holder of GREKT_TRUST_KEY can produce valid signatures — manually writing trusted: true in yaml is rejected.

  ✓ @grekt/tools        100  certified
  ⚠ @sketchy/tool        62  suspicious (trusted)

To remove trust after the author fixes the issues:

bash
grekt untrust @sketchy/tool

Result ​

Your workflow becomes:

  1. Discover an artifact you want to use
  2. Scan it for risks before installing
  3. Review any findings manually
  4. Install with confidence
  5. Enforce thresholds in CI so nothing slips through
  6. Trust reviewed exceptions without lowering the bar for everyone

Security doesn't slow you down. It runs alongside your normal workflow, individually and as a team.