dev-debug
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes the current codebase to identify root causes and implement fixes, which exposes the agent to untrusted data (source code and comments) that could contain malicious instructions.
- Ingestion points: Source code files read during the analysis phase in Step 1.
- Boundary markers: Absent. There are no instructions to the agent to treat code content as data only or to ignore instructions embedded in comments.
- Capability inventory: The skill has the ability to write to the file system (analysis reports and code fixes) and execute git commands.
- Sanitization: Absent. The skill does not validate or sanitize code content before using it to inform the fix implementation.
- [COMMAND_EXECUTION]: The skill performs shell-based operations to automate the version control process.
- Evidence: In Step 4, the skill executes
git add -A && git commit -m "fix: <concise description>". While the description is generated by the agent, if the agent's analysis is influenced by untrusted data to include shell metacharacters in the description, it could potentially lead to command injection depending on how the host platform handles shell execution.
Audit Metadata