The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Python scripts, validate-plan.py and validate-task-tracker.py, to enforce document schemas and logical consistency. Technical review of these scripts confirms they perform safe, string-based validation using standard libraries and do not possess networking or file-exfiltration capabilities.
[SAFE]: Operational safety is maintained through a strictly defined boundary between planning (read-only) and execution. The agent is prohibited from taking any irreversible or mutating actions until the user has explicitly reviewed and approved a 'READY_FOR_REVIEW' document.
[PROMPT_INJECTION]: The skill effectively mitigates injection risks by forcing all user inputs through a 'Bad Strategy Audit' and a structured 'Kernel' formulation. This process requires the agent to justify every action against a specific diagnosis and guiding policy, which helps prevent the execution of malicious instructions hidden in task descriptions.
[DATA_EXFILTRATION]: Analysis confirms the absence of exfiltration vectors. The skill does not access sensitive credential files, environment variables, or remote endpoints, and its execution environment is restricted to project-specific files using safe read-only tools.

plan-tasks