supply-chain-defense

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). These URLs include known IOC domains and exfil/C2 endpoints (webhook.site, blockchain-explorer APIs like api.ethplorer.io used by EtherHiding), raw IPs and obvious “evil” placeholders — all high-risk signals for malware C2/exfiltration or suspicious/untrusted download endpoints.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In scripts/integrity-audit.sh (AI-tool config persistence scan), the workflow reads user-/repo-local JSON/VS Code settings files and shell rc files (including colleague-/attacker-authored content that can be written into the monitored locations) as free text for grep/JSON parsing.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs the agent/user to add a remote depscore MCP server into Claude Code using the URL https://mcp.socket.dev/, which the assistant will query at runtime to obtain behavioural package scores that directly influence the agent's suggestions/decisions.