performance-review
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests performance data from several external and potentially attacker-controlled sources, which could contain malicious instructions designed to influence the agent's behavior during analysis.
- Ingestion points: Processes data from CSV exports, dashboard screenshots (OCR), and API responses from Google Search Console, GA4, PostHog, Beehiiv, and Kit (documented in SKILL.md, Phase 1).
- Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when interpolating this external data into the agent's context.
- Capability inventory: The skill has the ability to write to project files (
learnings.md,ad-matrix.json,assets.md) and perform further network requests based on its findings. - Sanitization: While the skill cleans data for formatting errors (e.g., abbreviations, locale decimals), it does not perform semantic sanitization to prevent instruction injection.
- [EXTERNAL_DOWNLOADS]: The skill performs legitimate network operations to well-known analytics and marketing services to retrieve report data.
- Service connections: Connects to
searchconsole.googleapis.com,analyticsdata.googleapis.com,us.posthog.com,api.beehiiv.com, andapi.kit.comusing provided API keys. - [COMMAND_EXECUTION]: The skill invokes local system components as part of its initialization and authentication flow.
- Script execution: Calls a local Python script
brand_context.pyto load environment-specific data. - CLI interaction: Provides instructions to use the
gcloudCLI tool for setting up application-default credentials with specific API scopes.
Audit Metadata