Comparison
Checked against each project’s README on 5 November 2026. These tools do different jobs and work well side by side.
| Tool | What it focuses on | Account or network | License |
|---|---|---|---|
| AgentShield | Claude Code setups (~/.claude/): secrets, permissions, hooks, MCP servers, prompt-injection vectors; --fix, GitHub Action, SARIF |
none for the scan; an optional deep-analysis mode calls Claude with your API key | MIT |
| snyk/agent-scan | MCP servers and skills across many agents: prompt injection, tool poisoning, malware in skills | needs a Snyk account and SNYK_TOKEN; starts or connects to MCP servers (with consent) to read their tools |
Apache-2.0 |
| agentsweep | secrets in agent history files across 32 agents, with in-place redaction | none, offline | MIT |
| gitleaks, trufflehog | secrets anywhere in git history and files; not specific to AI agents | none for gitleaks; trufflehog can verify keys with their providers | MIT; AGPL-3.0 |
| agentprint | keys, permissions, hooks and file modes across 9 agents’ files, plus agent files in npm/PyPI packages; installs itself into 10 agents as a pre-publish and pre-push check | none, offline; never uses a key it finds | MIT |