The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data by analyzing the entire repository, including source code, automation, configs, and documentation as defined in Step 1. It lacks explicit boundary markers to isolate this untrusted data from the agent's instructions. The agent has the capability to write files (Step 7) and monitor CI execution. While the skill requires the agent to verify all findings and provides security guidance for the generated output, the data ingestion phase remains a surface for indirect prompt injection.
[DATA_EXFILTRATION]: During repository analysis, the skill identifies necessary environment variables and secrets. It specifically instructs the agent not to execute commands that set secrets, instead providing guidance for the user, which prevents the exfiltration or exposure of credentials within the agent's execution context.
[COMMAND_EXECUTION]: The skill involves committing code and triggering workflows in Step 7 as part of its core functionality. It mitigates risk by recommending security practices for these workflows, such as disabling credential persistence and using environment protection.

generate-cicd