slate-patch
Warn
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill features an indirect prompt injection surface via the
$ARGUMENTSvariable and therepair <expectation>command. User-provided input in the prompt is directly used to patch rule files (.agents/rules/slate-patch.mdc), allowing for the persistent modification of the agent's behavioral instructions. - [COMMAND_EXECUTION]: The skill executes several shell commands and local scripts using both relative and absolute paths.
- It executes a script using a hardcoded absolute path:
/Users/zbeyens/git/plate-2/.agents/skills/autoreview/scripts/autoreview --mode local. Using absolute paths is a security risk as it assumes a specific local environment and could lead to the execution of unintended binaries if the path is compromised. - It runs
pnpm installto regenerate instructions after modifying rule files, which is a form of self-modifying behavior triggered by user input. - It executes local Node.js scripts:
node tooling/scripts/slate-autoresearch.mjs suggest-loops --with-checks.
Audit Metadata