debug-bolna-calls
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes raw call transcripts and logs retrieved from the Bolna API via the
diagnose_execution.pyscript. These logs contain untrusted user speech which could theoretically contain instructions designed to influence the AI agent's analysis. - Ingestion points:
scripts/diagnose_execution.pyfetching/executions/{id}/logand/executions/{id}. - Boundary markers: The script does not use specific delimiters or safety instructions when outputting the content of the logs to the agent's context.
- Capability inventory: The skill is capable of network reads and JSON parsing, and it provides a runbook for the agent to recommend changes to an agent's
task_configbased on the log data. - Sanitization: No sanitization of the transcript content or tool call payloads is performed before the data is presented to the agent for diagnosis.
- [COMMAND_EXECUTION]: The skill includes a Python script
scripts/diagnose_execution.pydesigned to be executed by the agent. This script performs network requests (urllib.request.urlopen) to the Bolna API (api.bolna.ai) to retrieve execution metadata and logs. This is expected functionality for the purpose of debugging the vendor's own voice agent platform.
Audit Metadata