debug-mode
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Node.js scripts to manage a background telemetry server. The collector binds only to loopback interfaces (127.0.0.1, localhost, or ::1) and utilizes secure file creation modes (0o700 for directories and 0o600 for files) to protect log data from other users on the system.
- [REMOTE_CODE_EXECUTION]: The agent is instructed to dynamically instrument workspace files with temporary code 'probes'. While this involves code generation and insertion, it is the primary intended function for diagnosing runtime bugs and is gated by manual user reproduction checkpoints.
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests and summarizes runtime logs from the application being debugged.
- Ingestion points: The
scripts/summarize-log.mjsutility reads NDJSON log files populated by instrumented application code. - Boundary markers: The instructions do not define specific delimiters to wrap log evidence in the summary provided to the agent.
- Capability inventory: The agent has the capability to modify project code and execute local scripts based on the resulting evidence summary.
- Sanitization: Log data is parsed as JSON, but the content of logged variables is processed as natural language evidence without escaping or filtering.
Audit Metadata