creating-backstage-plugin

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied descriptions and identifiers to scaffold new project files. This input is then interpolated into shell commands, creating a potential attack surface for command injection or directory traversal if an attacker provides a crafted plugin ID.
  • Ingestion points: User input captured in SKILL.md (Step 1: Capture intent) and references/templates.md (Step 8-9).
  • Boundary markers: The skill mentions a requirement for kebab-case identifiers, which serves as a weak boundary, but lacks explicit instructions to sanitize or validate the input against shell-specific metacharacters.
  • Capability inventory: The skill executes yarn new, ls, and yarn install via the shell (Step 4 and 5 in SKILL.md).
  • Sanitization: There is no explicit sanitization step described to prevent the user-supplied <id> from containing shell injection payloads like ; or &&.
  • [COMMAND_EXECUTION]: The skill involves the execution of shell commands (yarn new --select <template> --option pluginId=<id>) where the <id> is derived from user input. While intended for scaffolding, this mechanism can be exploited if the agent executes the string as a raw shell command without escaping.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 01:48 PM
Security Audit — agent-trust-hub — creating-backstage-plugin