The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and modify external marketing configurations such as Google Tag Manager, website tags, and API payloads. These sources can contain untrusted data with embedded malicious instructions designed to manipulate the agent's actions.
Ingestion points: As noted in references/skill-contract.md, the skill ingests data from public pages, exports, reviews, comments, and scraped materials.
Boundary markers: The instructions in references/skill-contract.md provide a defense by requiring that external evidence be treated as "untrusted input" rather than instructions.
Capability inventory: According to SKILL.md, the agent has the ability to implement authorized changes, publish tags, and edit accounts (e.g., "a tag publication, account edit, or upload can change records").
Sanitization: The skill contains explicit requirements in SKILL.md and references/verification.md to hash PII and ensure that sensitive data is not leaked into URLs, logs, or analytics events, which mitigates the risk of exfiltration via manipulated data.
[SAFE]: The skill provides links to official documentation from well-known services, including Google Analytics 4 and Meta Conversions API. These references are used for configuration guidance and do not represent a risk of remote code execution.

conversion-tracking