The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a learning system that processes task outcomes and contexts, which are ingestion points for untrusted data. This creates a surface for potential indirect prompt injection, where malicious data could influence future reasoning, although the impact is limited by the structured data format.
Ingestion points: rb.recordExperience and rb.learnPattern in SKILL.md accept runtime data.
Boundary markers: Not defined in the documentation.
Capability inventory: Limited to internal recommendation and database storage logic.
Sanitization: Not present in documented examples.
[SAFE]: The skill provides standard documentation and integration examples for a reasoning and learning library. All functionality aligns with the stated purpose of building self-learning agents. No unauthorized network activity or credential harvesting patterns were detected.

ReasoningBank Intelligence