lead-gen
Warn
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [CREDENTIALS_UNSAFE]: The script
scripts/research_leads.pyreads a Gemini API key from a local configuration file atreferences/credentials.md. Handling secrets in plain text files is a high-risk practice.\n- [PROMPT_INJECTION]: The skill is vulnerable to direct prompt injection. User-provided values forregionandcategoryare directly interpolated into the LLM prompt inscripts/research_leads.pywithin thebuild_promptfunction. A malicious user could provide input designed to override the agent's instructions and manipulate the AI's behavior.\n- [COMMAND_EXECUTION]: The skill's workflow inSKILL.mdinvolves executing local Python scripts using the Bash tool. These commands include variables like<region>and<category>that are derived from user input, which can be a vector for command injection if not properly handled by the agent environment.\n- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted data from the public web via Google Search grounding.\n - Ingestion points: Company information and contact details are retrieved from external websites by the Gemini API and passed back into the skill's logic in
scripts/research_leads.py.\n - Boundary markers: While the prompt requests data in a structured JSON format, there are no explicit boundary markers or instructions to the AI to ignore potentially malicious content found on the crawled websites.\n
- Capability inventory: The skill has the ability to write files to the user's filesystem (
generate_leads_xlsx.py) and execute shell commands (SKILL.md).\n - Sanitization: The skill parses the AI's response using
json.loads, but it does not perform textual sanitization or validation on the company data before it is written to the final Excel report.\n- [EXTERNAL_DOWNLOADS]: The skill references an external dependencyscripts/xlsx/recalc.pywhich is not provided within the skill's own directory. This represents an unverifiable script dependency that is required for certain optional features.
Audit Metadata