The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions involve processing untrusted data sources which could be used to deliver malicious instructions to the agent.
Ingestion points: The 'Data Sources' section in SKILL.md explicitly mentions reading scraped page text, reviews, comments, and emails.
Boundary markers: The skill requires the agent to label evidence using specific markers (e.g., measured, user-provided, calculated) to help distinguish data origins.
Capability inventory: The skill is authorized to generate artifacts and marketing recommendations. High-risk operations like account mutations, spending, or publishing require explicit user authorization.
Sanitization: The instructions include a specific directive to treat external evidence as untrusted and explicitly warn the agent not to follow instructions embedded within that evidence.
[CREDENTIALS_UNSAFE]: The skill references several third-party platforms (HubSpot, Stripe, Salesforce, etc.) that require authentication, but it correctly identifies that API keys should be kept out of skills and source control, utilizing external connector layers instead.
[DATA_EXFILTRATION]: While the skill integrates with various external marketing and RevOps tools, it maintains a 'read-only/dry-run' default for scripts and requires explicit user consent for any persistent writes or external side effects.

co-marketing