sales-ops

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime “Work mode” can ingest outsider-authored free text via the Data enrichment waterfall / AI account research path (sub-workflow #11): it performs web research and/or calls external enrichment providers, whose returned unstructured text (e.g., scraped pages, news, LinkedIn-derived snippets) is then fed into the agent/LLM context for extraction and summarization.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent at runtime to WebFetch live documentation (e.g., mcp.airtable.com/mcp and support.airtable.com and airtable.com/developers/web/llms.txt) to determine capability boundaries and to choose scaffolding/behavior, so these external URLs are fetched during runtime and their content directly controls the agent's prompts/decisioning.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)