process-infographic
Warn
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a bash script (
scripts/screenshot.sh) that executes a headless browser (Google Chrome or Chromium) to capture images. - Evidence: Step 6 in
SKILL.mdand the command execution logic inscripts/screenshot.sh. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by interpolating untrusted user data into an HTML context without sanitization.
- Ingestion points: User-provided process steps, titles, and descriptions identified in Step 1 of
SKILL.md. - Boundary markers: Absent; user content is placed directly into HTML tags.
- Capability inventory: Headless browser rendering with JavaScript support.
- Sanitization: Absent; the skill does not escape HTML entities or sanitize input before insertion into the template in Step 4.
- [DATA_EXFILTRATION]: The configuration of the headless browser creates a risk for local data exfiltration if malicious scripts are injected.
- Evidence:
scripts/screenshot.shuses the--no-sandboxflag, which disables a primary security layer in Chromium, and loads the generated HTML via thefile://protocol, which can grant access to the local file system.
Audit Metadata