The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration attempts were detected. The skill instructions prioritize user authorization and provide clear guidance on maintaining security boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates an attack surface related to processing untrusted external data (e.g., scraped web text, reviews, and emails). It includes specific defensive instructions to prevent the agent from following any malicious commands embedded in this data.
Ingestion points: External data sources including scraped page text, reviews, comments, emails, and third-party exports (documented in SKILL.md and references/skill-contract.md).
Boundary markers: Present; the skill explicitly directs the agent to treat these sources as untrusted evidence rather than instructions and to ignore any embedded commands.
Capability inventory: The skill can generate press materials and utilize connectors for search, SEO, and CRM tools, but all side effects and persistent writes require explicit user authorization.
Sanitization: The skill employs evidence labeling (e.g., measured, user-provided) and mandatory authorization checkpoints for external actions to ensure data integrity and user control.

public-relations