diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to ingest and act upon data from external, untrusted sources.\n
- Ingestion points: The skill instructions in SKILL.md direct the agent to read external files such as CONTEXT.md, ADRs, log files, network traces (HAR files), and command outputs to build a mental model and diagnostic hypothesis.\n
- Capability inventory: The agent is empowered to execute shell commands, run bash scripts (e.g., scripts/hitl-loop.template.sh), and invoke external tools like curl or headless browsers.\n
- Boundary markers: There are no instructions provided to use delimiters or ignore potential instructions embedded within the logs or trace data being analyzed.\n
- Sanitization: The workflow lacks steps for sanitizing, validating, or filtering the content of external files before the agent processes them.\n- [COMMAND_EXECUTION]: The methodology relies on the agent creating and executing diagnostic loops, which involves running shell commands, bash scripts, and network tools (curl) in the local environment based on findings from the codebase.
Audit Metadata