connect-agent
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
langwatchCLI to register agents, manage secrets, and access documentation within the environment. - [EXTERNAL_DOWNLOADS]: Documentation is retrieved from vendor-controlled domains (
langwatch.ai). The reporting utility is executed vianpx(which pulls thelangwatchpackage from the registry). - [DATA_EXFILTRATION]: A reporting mechanism (
npx langwatch report) allows the agent to send session transcripts and summaries to the LangWatch team for feedback. This action is restricted by a--user-approvedflag and includes local scrubbing logic to remove common secrets before transmission. - [PROMPT_INJECTION]: The creation of a new HTTP endpoint for agent interaction introduces a surface area for indirect prompt injection via incoming simulation traffic. 1. Ingestion points: Agent HTTP endpoint created/modified in Step 2. 2. Boundary markers: Absent in provided configuration snippets. 3. Capability inventory: CLI execution, codebase modification, and environment variable access. 4. Sanitization: Absent in provided snippets.
- [CREDENTIALS_UNSAFE]: The skill directs the agent to locate and read the
LANGWATCH_API_KEYfrom a.envfile. It includes specific constraints to ensure the agent does not print, copy, or exfiltrate this key.
Audit Metadata