Investigation

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill requires the mandatory execution of a background shell command (curl) for local status notifications (targeting localhost:8888) across all investigation workflows.
  • Evidence: Multiple workflow files (e.g., OSINT/Workflows/PeopleLookup.md, PrivateInvestigator/Workflows/FindPerson.md) contain a 'MANDATORY: Voice Notification' section requiring curl -s -X POST http://localhost:8888/notify ....
  • Vulnerability: The instruction directs the agent to interpolate variables like WORKFLOWNAME and ACTION into the shell command string, which creates a potential command injection risk if user-supplied data is used for these fields without sufficient sanitization.
  • [PROMPT_INJECTION]: The skill includes instructions to load and apply external configurations and preferences from the local filesystem, which can override default behavior.
  • Evidence: Both OSINT/SKILL.md and PrivateInvestigator/SKILL.md contain a 'Customization' section instructing the agent to check ~/.claude/PAI/USER/SKILLCUSTOMIZATIONS/ and load PREFERENCES.md or other resources which 'override default behavior'.
  • [PROMPT_INJECTION]: The skill exhibits a significant surface for Indirect Prompt Injection through its use of parallel research agents to gather data from the web.
  • Ingestion points: Untrusted data enters the context from multiple external research agents (Perplexity, Grok, Gemini, Claude) fetching web search results, social media profiles, and public records.
  • Boundary markers: Absent. The prompts for sub-agents (e.g., in OSINT/Workflows/PeopleLookup.md) do not include delimiters or instructions to ignore embedded commands within the searched content.
  • Capability inventory: The agent has access to shell command execution (curl), file reading and writing (specifically in the DiscoverOSINTSources workflow), and sub-agent delegation.
  • Sanitization: Absent. There are no instructions for the agent to sanitize or escape the content returned by research tasks before synthesis.
  • [EXTERNAL_DOWNLOADS]: The documentation references and provides instructions for installing various third-party OSINT tools and libraries from public registries.
  • Evidence: OSINT/PeopleTools.md and PrivateInvestigator/Workflows/ReverseLookup.md mention pip install holehe, pip install sherlock-project, and npm install -g node-ipinfo.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 01:52 AM