plan-design
Warn
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
curlwithin a bash script to interact with the OpenAI Images API. Variables like$PROMPT_A,$PROMPT_B, and[surface-slug]are interpolated directly into shell commands. If these values contain shell metacharacters (e.g., through a maliciously crafted Linear issue title or surface name), it could lead to arbitrary command execution on the host machine. - [EXTERNAL_DOWNLOADS]: The skill downloads image assets from OpenAI's API (
api.openai.com) and saves them locally usingbase64 -d. While OpenAI is a well-known service, the skill retrieves this data over the network and executes a decoding utility on the output. - [REMOTE_CODE_EXECUTION]: The skill uses a
Tasktool to spawn a subagent that generates HTML/CSS files. While the subagent itself is part of the agent's internal capabilities, the generated HTML is written to the local filesystem and subsequently uploaded to Linear, creating a path where generated code is stored and distributed.
Audit Metadata