Agent Guard module

Agent supply chain security for skills, MCP, and IDE rules

Static package scanners miss natural-language trojans. Agent Guard reads skill instructions, MCP JSON, and Cursor rule files for exfiltration endpoints, typosquatted registries, and IOCs from active 2026 campaigns including Paperclip-style skill attacks.

• Paste a skill file or MCP config and get a block, warn, or pass verdict in seconds.
• Matches curated IOCs: malicious domains, npm typosquats, and known campaign markers.
• REST at POST /v1/agent-scan and MCP tool scan_agent_artifact for CI and agent workflows.
• Findings map to OWASP, CWE, SOC 2 CC, and CIS v8 for audit packets.

Scan an agent artifact API docs

Why agent supply chain is a 2026 priority

Developers install skills from marketplaces and wire MCP servers without the same review rigor applied to npm packages. Attackers embed callback URLs and credential harvesters inside markdown instructions that never touch a dependency graph. Security teams need a reader that understands agent artifacts, not only lockfiles.

Noetfield Shield Agent Guard complements SAST and dependency scanners. It is deterministic, rate-limited, and does not store uploaded content beyond optional shareable reports you explicitly request.