caveman-compress
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python module from the shell using the command
python3 -m scripts <absolute_filepath>. This command runs a script located in an adjacent directory to the skill definition. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from user-specified text files (.md, .txt, etc.) and passes this content to the LLM for compression.
- Ingestion points: Arbitrary text files provided via the
<absolute_filepath>argument. - Boundary markers: None are defined in the instructions to separate the untrusted file content from the compression instructions.
- Capability inventory: The skill has the capability to read files, write files (backups and overwrites), and execute shell commands.
- Sanitization: There is no mention of sanitizing or escaping the content of the processed files before they are sent to the LLM for compression.
Audit Metadata