The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized access attempts were detected. The skill is primarily instructional and contains multiple security guardrails, such as mandatory authorization for side effects and clear labeling of evidence uncertainty.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting untrusted data from scraped web content, emails, reviews, and comments (SKILL.md, references/skill-contract.md). This risk is addressed through explicit boundary instructions that mandate treating such input as data rather than instructions, and a requirement to ignore any commands embedded within evidence.
[DATA_EXFILTRATION]: The skill defines connectors for interacting with high-value marketing and financial platforms including Stripe, HubSpot, Google Ads, and various ESPs (references/connectors.md). To prevent unauthorized data movement or account changes, the skill contract enforces a requirement for explicit user authorization before any external side effects, spending, or persistent writes are executed.

ab-testing