media-pipeline-router

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a dispatch mechanism that builds subagent prompts by concatenating system-level instructions with user input.
  • Ingestion points: The orchestrator reads markdown files from the local disk at ${CLAUDE_PLUGIN_ROOT}/modes/_shared.md and mode-specific files like ${CLAUDE_PLUGIN_ROOT}/modes/{mode}.md (SKILL.md).
  • Boundary markers: Delimiters between the system-defined instructions and the user's "exact ask" consist only of two newlines. No robust markers (e.g., XML tags) or security headers are used to ensure the subagent ignores instructions embedded within the user's input (SKILL.md).
  • Capability inventory: Spawned subagents have significant capabilities, including access to "every Bash CLI" and specialized tools like moprobe, moqc, and mosafe (SKILL.md).
  • Sanitization: The instructions explicitly mandate using the user's "exact ask" and forbid paraphrasing, with no sanitization or escaping performed before concatenation (SKILL.md).
  • [COMMAND_EXECUTION]: The orchestrator is designed to facilitate the execution of shell commands and media processing utilities via its specialist subagents. It describes a system where agents have broad access to CLI tools to perform tasks (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 09:54 AM
Security Audit — agent-trust-hub — media-pipeline-router