seo-intel
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites to perform technical audits and content gap analysis.
- Ingestion points: Data enters the agent's context through
nimble extractandnimble crawlcommands targeting user-specified URLs, as seen in the workflows for site audits (references/wf-site-audit.md) and competitor keywords (references/wf-competitor-keywords.md). - Boundary markers: The skill employs structured sub-agent prompts (e.g.,
references/seo-extraction-prompt.md) which explicitly instruct agents to "Extract and analyze" and "No analysis, no interpretation," designed to limit the agent's focus to data extraction and prevent the execution of instructions found in the source content. - Capability inventory: The skill utilizes
Bashfor command execution,Agentfor spawning parallel sub-agents, andWritefor local data persistence. These capabilities represent the potential impact surface if an injection bypasses the provided sub-agent instructions. - Sanitization: The skill relies on structured JSON extraction and LLM-based parsing (
--parse --parser) but does not include explicit sanitization of the extracted site text before processing. - [COMMAND_EXECUTION]: The skill utilizes the
Bashtool to perform its core operations via CLI wrappers. - Evidence: Multiple reference files (e.g.,
references/wf-rank-tracker.md,references/wf-ai-visibility.md) construct and execute shell commands using thenimbleandghCLIs based on user-provided domain and keyword arguments. - [DATA_EXFILTRATION]: The skill includes functionality to move researched data to external services.
- Evidence: As detailed in
references/memory-and-distribution.md, the skill facilitates report distribution to Notion and Slack via their respective MCP connectors. This represents a legitimate data flow from audited sites to user-controlled collaborative platforms. - [EXTERNAL_DOWNLOADS]: The skill fetches data and interacts with multiple well-known third-party services.
- Evidence: The skill communicates with GitHub via the
ghAPI and queries AI search platforms including ChatGPT, Perplexity, Gemini, and Grok using Nimble agents to gather visibility metrics.
Audit Metadata