The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions in references/audit-blog.md direct the agent to 'Read the curl command from the response' of a tool and then 'Run the returned curl command.' This pattern of executing instructions directly from an external tool's output constitutes dynamic code execution and poses a risk of remote code execution if the tool returns a malicious payload.
[COMMAND_EXECUTION]: The skill uses the shell to perform network operations. Specifically, in references/audit-blog.md, the agent is instructed to run curl commands to upload local draft content to external URLs. This involves assembling shell commands with variables (such as UPLOAD_URL) derived from tool responses.
[DATA_EXFILTRATION]: In references/audit-blog.md, the agent is instructed to use curl to upload the contents of the local file {PROJECT_ROOT}/blogs/<topic>/blog.md to an external URL provided by the audit_content_quality tool. While this is part of the intended audit workflow, it technically performs data exfiltration of local user content to the network.

blog-creator