deepline-feedback
Warn
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local system commands using the
deeplineCLI tool. As defined in Step 3 of SKILL.md, it runsdeepline provide-feedbackanddeepline session send. - [DATA_EXFILTRATION]: The skill is designed to export the current session transcript and environment information to an external team. While Step 2 includes a user confirmation prompt, the automated collection and transmission of full session history represents a significant data exposure surface if handled improperly by the receiver.
- [PROMPT_INJECTION]: User-supplied feedback text is directly interpolated into a shell command in SKILL.md:
deepline provide-feedback --text "{feedback text}". There are no instructions for the agent to sanitize the input, which could allow an attacker to use shell metacharacters (e.g., backticks or semicolons) to escape the intended command and execute unauthorized actions. - [REMOTE_CODE_EXECUTION]: The vulnerability to command injection in the feedback processing step could be exploited to achieve arbitrary code execution on the system where the agent is running.
Audit Metadata