gs-scratch
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs shell operations and file manipulations that could be exploited for path traversal.
- The
writeandeditcommands inSKILL.mdinterpolate a user-provided<filename>into the path.scratch/<filename>. - Without explicit instructions to the agent to sanitize or validate this filename, there is a risk that path traversal sequences (e.g.,
../) could be used to write to or edit files outside the.scratchdirectory. - [INDIRECT_PROMPT_INJECTION]: The skill reads the contents of existing files in the scratch directory, which is a common vector for indirect prompt injection.
- Ingestion points: The
editcommand (as described inSKILL.md) reads the contents of.scratch/<filename>into the agent's context. - Boundary markers: The instructions do not specify any boundary markers or ignore-rules to protect against instructions embedded within the file content.
- Capability inventory: The skill has the ability to execute shell commands (
mkdir,ls) and perform file system modifications using the agent's toolset. - Sanitization: There is no mentioned sanitization of the file content before it is read and processed by the agent.
Audit Metadata