langfuse-debug-bundle
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_CONTEXT_INJECTION
Full Analysis
- [DYNAMIC_CONTEXT_INJECTION]: The skill uses dynamic context injection to verify and display the versions of Node.js and Python, and check for installed Langfuse packages during the skill initialization phase.
- [DATA_EXFILTRATION]: The skill accesses sensitive environment variables ('LANGFUSE_SECRET_KEY') and application log files. It performs an active network test by sending credentials in a Base64-encoded 'Authorization' header to the Langfuse API host to verify health and connectivity.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by reading application logs from the 'logs/' directory. This data is included in the diagnostic bundle without robust sanitization beyond API key redaction, creating a surface for indirect prompt injection.
- Ingestion points: Application logs ('logs/*.log').
- Boundary markers: None used to delimit log content within the script's output.
- Capability inventory: The skill utilizes 'Bash' with 'curl', 'grep', and 'tar' for data gathering and exfiltration.
- Sanitization: Uses 'sed' to redact specific API key prefixes but does not filter for potential malicious instructions in logs.
- [COMMAND_EXECUTION]: The skill generates and facilitates the execution of a diagnostic shell script that utilizes system tools like 'curl' and 'tar' to aggregate and transmit system information.
Audit Metadata