audit-langfuse-llm

Warn

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructs the agent to search for and read sensitive environment variables, specifically LANGFUSE_SECRET_KEY, from .env files. While the instructions advise against recording the secret value in the final report, the execution of the Grep command brings the plaintext secret into the active agent context.
  • [REMOTE_CODE_EXECUTION]: The skill downloads and executes CLI tools, specifically langfuse-cli and @playwright/cli, using npx. These tools are fetched from public registries at runtime to perform audit and browser automation tasks.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by scraping external web content via Firecrawl and ingesting Langfuse trace data to inform its auditing logic.
  • Ingestion points: Web search results and scraped markdown content from Phase 1a and 1b; Langfuse trace data from Phase 2.
  • Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious instructions embedded in the scraped web content.
  • Capability inventory: The skill possesses high-privilege capabilities including shell command execution via npx, sensitive file system access via Grep, and database interaction via Supabase tools.
  • Sanitization: No sanitization or validation of the external content retrieved via Firecrawl is mentioned before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands like Grep and npx to interact with the local filesystem and execute external packages, which are essential for its stated purpose of auditing the codebase and integration.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 7, 2026, 11:26 PM