The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a local Python utility at .claude/scripts/code_graph to perform code analysis tasks such as tracing call flows and mapping dependencies. The agent is instructed to run this script using python with various JSON-output flags.
[PROMPT_INJECTION]: The skill incorporates an injection surface where user-provided input is placed into the $ARGUMENTS variable inside <tasks> tags. While this allows the user to define the work to be done, it also allows arbitrary text to be inserted into the agent's immediate instruction stream.
[SAFE]: The skill demonstrates positive security posture by implementing 'Red Flag Stop Conditions' which explicitly block the agent from proceeding without user approval when handling security-sensitive code involving authentication, cryptography, or PII. It also mandates reading existing code and verifying patterns before making any modifications, reducing the risk of accidental vulnerability introduction.

cook-fast