canonical-format-checker

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill directs the agent to read and follow "authoritative format" instructions from local files (e.g., .claude/skills/, .claude/agents/). This creates a vulnerability where an attacker who can modify these files (e.g., via a Pull Request) can inject malicious instructions that the agent will treat as mandatory "Canonical Format" rules during content creation or review.
  • Ingestion points: File reading via cat and head in SKILL.md (Step 2).
  • Boundary markers: Absent. The agent is not instructed to treat the content as untrusted data.
  • Capability inventory: File system inspection (ls) and reading (cat, head).
  • Sanitization: Absent. Content is read directly into context for comparison.
  • [Command Execution] (LOW): The skill explicitly instructs the agent to execute shell commands (ls, cat, head) to inspect the local filesystem. While these are restricted to read-only operations in the provided examples, they provide the agent with a mechanism to access sensitive internal configuration directories.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 05:57 AM