The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads local project documentation, including README, DESIGN.md, and ADRs, to gather facts. While these files represent an external data ingestion surface that could theoretically contain malicious instructions, the skill's design limits the impact by only using the information to formulate questions for the user. Furthermore, it explicitly requires user confirmation before any implementation or action is taken.
[SAFE]: The skill's logic is transparent and focuses on structured user interaction. No patterns of obfuscation, remote code execution, or unauthorized data access were identified. It adheres to safety principles by instructing the agent to wait for user confirmation before acting on the gathered information.

w-grill