The Agent Skills Directory

[SAFE]: The skill is designed to prevent autonomous agents from making unreviewed or destructive changes by enforcing a human-in-the-loop planning gate.
[PROMPT_INJECTION]: There are no instructions that attempt to override safety filters or the agent's core mission. The instructions focus on establishing a methodical and evidence-based workflow.
[DATA_EXFILTRATION]: The protocol focuses on reading project context for analysis and does not include any commands for exfiltrating sensitive files or credentials.
[REMOTE_CODE_EXECUTION]: The skill explicitly forbids actions like installing packages, updating lockfiles, or running migrations during the planning phase, effectively preventing RCE through the planning channel.
[COMMAND_EXECUTION]: The skill restricts tool usage to read-only or non-mutating diagnostic commands, ensuring that project state is preserved during investigation.
[EXTERNAL_DOWNLOADS]: No references to external scripts, remote downloads, or untrusted third-party dependencies were found.

plan-mode