The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong defensive instructions in its 'Hard Policy' section, explicitly stating that user-supplied phrases cannot override the planning boundaries. This protects against attempts to force the agent into executing code prematurely.
[COMMAND_EXECUTION]: The skill invokes local utility scripts (e.g., create-goal-scratchpad.mjs, check-complete.mjs) and standard development tools like rg (ripgrep) and pnpm. These operations are used for internal state management and verification within the project's own directory structure.
[SAFE]: The skill implements a 'Read First' policy and license/provenance discipline that requires the agent to summarize behavior rather than copying external code or prose, which effectively mitigates risks from potentially malicious data in processed reports.

editor-harvest-plan