The Agent Skills Directory

[SAFE]: The skill implements a legitimate and helpful cognitive pattern for AI agents, focusing on long-term memory and learning through reflection and persistent storage.
[PROMPT_INJECTION]: The skill involves capturing and later reviewing data from past interactions, which could theoretically include content from untrusted sources (e.g., tool outputs or external code comments). However, the process includes mandatory reflection and distillation steps. Crucially, the skill explicitly requires user authorization before promoting any captured pattern to a permanent project rule (CLAUDE.md), providing a strong human-in-the-loop control against indirect prompt injection.

knowledge-capture