LocalIntelligence

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local TypeScript tools using the bun runtime for daily data fetching, history backfilling, and content processing.
  • [COMMAND_EXECUTION]: In Tools/ClaudeFill.ts, the skill spawns a claude subprocess to perform targeted web research using WebSearch and WebFetch tools. The subprocess environment is configured to use the user's existing session by unsetting environment variables like ANTHROPIC_API_KEY to prevent conflicts.
  • [DATA_EXFILTRATION]: The skill accesses the local file ~/.claude/LIFEOS/USER/PRINCIPAL/PRINCIPAL_IDENTITY.md to retrieve user location data (city, state, zip). This data is used to scope local searches and is included in search queries sent to the web research sub-agent.
  • [EXTERNAL_DOWNLOADS]: The skill performs network requests to fetch RSS feeds from patch.com and other user-provided URLs defined in sources.json. It also utilizes a sub-agent to fetch information from search results.
  • [PROMPT_INJECTION]: The skill processes untrusted external data from news feeds and web search results. It maintains an evidence chain where ingestion occurs through external APIs and searches, boundary markers are defined by strict JSON output requirements in sub-agent prompts, capabilities include file writing and tool execution, and sanitization is enforced by the validateSection function which performs schema and URL validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 06:48 PM
Security Audit — agent-trust-hub — LocalIntelligence