tweak
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes shell commands (
start,open, orxdg-open) to launch local HTML files in the user's default browser for interactive tweaking. It also provides specific instructions for the agent to use command-line utilities such asgrepandsedto identify marker strings and safely delete line ranges when stripping the injected panel from the file.- [INDIRECT_PROMPT_INJECTION]: The skill features a capability to ingestion data generated outside the agent context via the 'Bake' workflow. The agent processes a JSON payload (BAKE_TWEAK_v1) to update hardcoded CSS values in local files. * Ingestion points: User-supplied JSON payload pasted from a browser session into the chat context. * Boundary markers: Explicit header stringBAKE_TWEAK_v1and JSON object structure. * Capability inventory: File system modification (HTML/CSS) and shell execution. * Sanitization: The instructions guide the agent to perform deterministic updates to CSS rules based on values provided in the JSON, rather than executing arbitrary instructions contained within the data.
Audit Metadata