seo-geo
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documentation instructs the agent to utilize local Python scripts (
manage_markdown_site.pyandmanage_markdown_work.py) for site generation, validation, and workflow management. These scripts are part of the vendor's provided suite and are executed within the local environment. - [INDIRECT_PROMPT_INJECTION]: The skill processes external web captures and research data which are potentially attacker-controlled. However, the skill implements multiple layers of protection: it requires verification of immutable hashes for all captures and metadata, and validates all structured inputs against strict JSON schemas before analysis.
- Ingestion points:
research-pack.jsonand raw web captures (HTML/content) located inraw/observations/. - Boundary markers: The protocol mandates verification of cryptographic hashes and stable-ID uniqueness before any analysis begins.
- Capability inventory: The agent performs file system writes for reports (e.g.,
geo-findings.md) and executes local suite-management scripts. - Sanitization: Inputs must conform to schemas like
research-pack.schema.json, and the skill specifically instructs the agent to decline speculative or unverified claims.
Audit Metadata