deepline-feedback
Warn
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: In Step 3 of SKILL.md, the instruction directs the agent to execute
deepline feedback send "{feedback text}" --json. This is vulnerable to shell command injection because user-supplied text inside double quotes can still trigger command substitution (e.g., via$(...)or backticks) or escape the quotes entirely. - [DATA_EXFILTRATION]: The skill is designed to collect and upload sensitive data, specifically the environment info and the current session transcript, to an external service. While it mandates user confirmation via
AskUserQuestion, the exfiltration of entire conversation histories represents a significant privacy risk. - [COMMAND_EXECUTION]: The skill uses a sequence of shell commands in Step 4 (
mkdir,ln,ls,deepline sessions send) to programmatically locate and link session files. These commands are executed based on local file system state and glob patterns, which could lead to unexpected behavior if the environment is manipulated.
Audit Metadata