The Agent Skills Directory

[DATA_EXFILTRATION]: Automated security scans (URLite) have identified the URL https://oauth-provider.com/token as malicious and blacklisted. This URL is included in SKILL.md under the troubleshooting section (Error #24) for OAuth remote connection failures, where users are explicitly instructed to use it in a fetch command to verify connectivity. If a developer follows these instructions while configuring their own authentication flow, sensitive credentials or session tokens could be exfiltrated to this malicious domain.\n- [INDIRECT_PROMPT_INJECTION]: The commands/init.md file defines a workflow that ingests untrusted user input (e.g., project names) which is then used directly in shell commands such as npm create. The skill lacks boundary markers or sanitization logic to prevent malicious inputs from performing command injection or influencing the agent's behavior during project initialization.\n- [COMMAND_EXECUTION]: The skill requests high-privilege tool access (including Bash, Write, and Edit) and relies on executing shell commands for its core functionality. While expected for a developer tool, the combination of high-privilege access and a confirmed malicious URL hit significantly increases the risk profile.\n- [METADATA_POISONING]: The skill contains deceptive safety claims in SKILL.md, asserting a '100% prevention rate' of errors and a 'Production Ready' status. These authoritative claims are contradicted by the automated scanner findings (FileRepMalware) and the inclusion of blacklisted infrastructure. Additionally, there is an identity discrepancy between the reported author (ma1orek) and the metadata (Jeremy Dawes).

cloudflare-mcp-server