ucoz-ad-campaign-landing-skill
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands (curl) extensively in
references/ADS.mdandreferences/GOOGLE_ADS.mdto interact with official Yandex Direct and Google Ads APIs for campaign management, research, and link synchronization.\n- [DYNAMIC_EXECUTION]: The skill contains an embedded PowerShell script inreferences/LANDING_PLAYBOOK.md§4.2 for programmatically generating favicon files. It also uses Python one-liners to parse JSON data from API responses inreferences/ADS.md. These are functional requirements for the skill's primary purpose of landing page and asset creation.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection, as it ingests untrusted data from external sources including CSV files (templates/ads.example.csv), Yandex Search API results, and existing uCoz page content. \n - Ingestion points: CSV ad templates, search engine result pages (SERP), and uCoz template data.\n
- Boundary markers: The instructions lack explicit delimiters for separating external ad copy from system instructions, relying instead on structural validation.\n
- Capability inventory: The skill can perform file writes (
files_write), template updates (templates_write), and network API calls.\n - Sanitization: The skill mitigates risks through mandatory 'Message Match' overlap calculations, automated template validation (
validate_template), and manual human review checkpoints before any production launch.\n- [DATA_EXFILTRATION]: The skill includes a platform-specific usage registration call (skills_tool) at the end of its workflow to track execution metrics. It also accesses account-level data from Yandex and Google APIs to generate reports for the user.
Audit Metadata