The Agent Skills Directory

[SAFE]: The skill manages project state locally using a dedicated factory/ directory, keeping project data within the local environment.
[SAFE]: It coordinates between specific sub-skills (factory-plan, factory-tests, etc.) that must be locally installed, avoiding the risks associated with remote code execution.
[SAFE]: Human-in-the-loop design is strictly enforced; the skill provides instructions for the user to launch build cycles rather than executing them autonomously.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it reads and interprets build logs (factory/log.md) and progress reports (factory/progress.md). If these files (typically generated by automated processes) were to contain malicious instructions, they could influence the agent's summary and subsequent stage transitions. Ingestion points: File reads from factory/progress.md and factory/log.md during stage 5 status checks. Boundary markers: None mentioned for log file ingestion. Capability inventory: Local file system read/write and invocation of sibling skills. Sanitization: No validation or sanitization is mentioned for the content of processed log files.

factory