skills/autumnsgrove/lattice/cat-check/Gen Agent Trust Hub

cat-check

Warn

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using uv run to interact with project tools like glimpse. It directly interpolates user-supplied variables such as {target_url}, {tenant}, and {page-name} into these commands. This creates a potential vector for command injection if the user provides malicious payloads instead of valid URLs or names.\n- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves capturing screenshots of external websites and analyzing them for 'warmth' and 'brand vibes'. This ingestion of untrusted visual data represents an indirect prompt injection surface where a malicious page could include text designed to manipulate the agent's assessment or trigger unintended behavior.\n
  • Ingestion points: External web page content captured via glimpse screenshots as described in SKILL.md.\n
  • Boundary markers: Absent; there are no delimiters or specific instructions for the agent to treat the screenshot content as purely passive data.\n
  • Capability inventory: Shell command execution via uv run, local file system reads (design references and UI components), and report generation.\n
  • Sanitization: Absent; the skill does not define methods to sanitize or filter the content of captured screenshots before the agent processes them.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch and capture arbitrary external URLs provided by the user via the glimpse matrix and glimpse capture commands.\n- [REMEDIATION]: To improve security, user-supplied variables should be strictly validated and sanitized before being used in shell commands. The agent should also be explicitly instructed to ignore any text found within screenshots that appears to be a command or instruction, treating the visual data as a subject of analysis only.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 9, 2026, 01:31 AM