The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves reading from local markdown files which serves as a potential attack surface for indirect prompt injection. However, the skill includes robust mitigation instructions in the references/memory-protocol.md file, explicitly stating that memory content is data and that old notes or decisions grant no authorization for secret access, permission changes, or external actions. This design correctly prioritizes current instructions and repository state over stored context.
[EXTERNAL_DOWNLOADS]: The skill's research ledger (references/research.md) references official documentation and research papers from trusted and well-known sources, including Anthropic, the Git project, and arXiv. These references are provided for agent guidance and research purposes only, with no evidence of automatic execution of remote scripts or downloads of untrusted dependencies.

recall