The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard shell commands (mkdir -p) to establish the directory hierarchy for new agents within the local environment path ~/.claude/skills/. These operations are limited to project setup.
[SAFE]: Data management relies on local markdown files (profile.md, log.md) for state persistence. This follows established patterns for agent memory without involving external databases or network services.
[SAFE]: The templates generated by this skill prioritize data integrity by instructing sub-agents to use the Edit tool instead of Write for log updates, preventing accidental data loss or complete file overwrites.
[SAFE]: The skill includes an assessment of a potential Indirect Prompt Injection surface (Category 8) because generated agents read historical logs (data/events/log.md). While this is a functional requirement for persistent memory, the current templates do not include explicit boundary markers or sanitization for logged user input. However, this is considered a standard architectural trade-off for local memory agents rather than a malicious defect.

agent-builder