The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to perform standard local development tasks, such as running project tests using go test, package.json scripts, or make/just targets. These are executed within the local environment for verification purposes.
[DATA_EXFILTRATION]: Operations that could move data externally, such as push or opening a Pull Request, are explicitly set to be disabled by default. The instructions mandate that these actions occur only upon an explicit, separate request from the user.
[PROMPT_INJECTION]: The workflow incorporates a 'Confirm gate' (hard stop) which requires the user to give an explicit go-ahead before the agent can begin writing code. This prevents the agent from autonomously pursuing tasks without human oversight.
[SAFE]: The skill uses a 'Freeze' step that involves an audit process to review all decisions and changes before they are committed. This multi-stage verification process aligns with security best practices for agent-led development.

develop