The Agent Skills Directory

[COMMAND_EXECUTION]: The skill describes an exec: verb used for executing shell commands. This capability is managed through a 'Native-first' policy and a mandatory 'Exec ledger' for documentation, which are design patterns intended to minimize the use of arbitrary shell execution in favor of restricted built-in tools.
[EXTERNAL_DOWNLOADS]: Mentions the use of MCP (Model Context Protocol) tools and a built-in nika:fetch tool for HTTP operations. These are standard features for a workflow automation engine and are documented within a security-conscious framework that includes cost and permission auditing.
[CREDENTIALS_UNSAFE]: The instructions explicitly guide the user to handle sensitive data safely using a ${{ secrets.X }} syntax and environment variables, rather than hardcoding credentials. It also describes secret masking and egress sink declarations as part of the tool's standard operation.
[DATA_EXFILTRATION]: While the tool has network and file access capabilities, the instructions focus on auditability and restricted permissions (permits: blocks), providing a framework for safe data handling rather than demonstrating malicious exfiltration patterns.

nika