The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were identified. The skill is exceptionally well-documented and follows industry best practices for advertising account management and data processing.
[INDIRECT_PROMPT_INJECTION]: The skill contains a potential surface for indirect prompt injection through its brand analysis feature, which is a standard functional requirement for this use case.
Ingestion points: The /ads dna <url> command (SKILL.md) crawls user-provided websites to extract brand attributes into a brand-profile.json file.
Boundary markers: While the orchestration logic in SKILL.md specifies that the agent must "validate each subagent returned valid JSON scores," there are no specific delimiters or escaping mechanisms described for the text extracted from external websites.
Capability inventory: The skill possesses significant capabilities, including file system writes (brand-profile.json, campaign-brief.md), creative copy generation, and automated image generation via external APIs.
Sanitization: The skill focuses on structural validation of JSON data but does not explicitly detail the sanitization of natural language instructions that might be embedded in crawled website metadata.
[COMMAND_EXECUTION]: The skill uses a defined orchestration logic to spawn sub-agents (e.g., audit-google, audit-meta) via a Task tool with context: fork. This is a standard platform capability used for parallel processing and does not constitute unauthorized command execution.
[CREDENTIALS_UNSAFE]: The skill correctly handles sensitive credentials by requiring them to be provided via environment variables (e.g., GOOGLE_API_KEY, OPENAI_API_KEY) rather than hardcoding them within the skill files.

ads