The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes two helper scripts, scripts/task-start and scripts/task-done, which use standard Git and shell commands to manage a development workflow. They interact with a companion skill (subagent-driven-development) to maintain workspace isolation and log progress. These operations are restricted to the local repository and the .superpowers/ directory.
[DATA_EXPOSURE]: The skill implements a 'ledger' system (progress.md) to track task status across session compactions. This data remains local to the user's workspace and is used for consistency rather than exfiltration.
[PROMPT_INJECTION]: The instructions contain phrases like 'The plan already did the thinking. Execute it exactly,' and 'Continuous execution: Do not pause.' While these are directive, they are within the context of following a user-approved plan and are balanced by safety constraints (e.g., stopping for destructive operations or security-sensitive actions).

executing-plans