The Agent Skills Directory

[SAFE]: The skill implements a secure execution protocol using 'blocking' markers ([STOP — BLOCKING]). These markers require explicit user intervention and confirmation before the agent can proceed with decision-making or file operations, ensuring the user maintains control over the output.\n- [SAFE]: Local project context ingestion is scoped to documentation directories (docs/product and docs/discovery) for reading and restricted to a specific design documentation path (docs/product/design/) for writing. No patterns of unauthorized data exfiltration or sensitive file access were identified.\n- [SAFE]: The use of external tools, such as the Agent tool for sub-agent delegation and WebSearch for brand research, is functionally aligned with the skill's purpose. All findings from these tools are subject to user review during the mandatory confirmation phases.

recipe-blueprint