The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it is designed to ingest untrusted data from external sources including scraped web text, reviews, comments, and emails (specified in SKILL.md and references/skill-contract.md). However, it proactively mitigates this risk through explicit instructions to treat such data as evidence only and to ignore any embedded instructions. The skill contract further restricts capabilities by requiring explicit authorization for any persistent writes or real-world side effects, ensuring that even if an injection were successful, the agent's actions remain constrained.
[SAFE]: The skill architecture adheres to high security standards, incorporating a 'Skill Contract' that enforces least privilege and user consent. It explicitly forbids the hardcoding of API keys, mandates authorization for any account-mutating actions, and requires clear evidence labeling (measured vs. estimated) to prevent the upgrade of uncertainty. No malicious code, obfuscation, or unauthorized network operations were detected in the provided files.

offers