file-review

Pass

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes feedback extracted from markdown files using specific HTML markers (e.g., <!-- review-end(ID): feedback -->). This feedback is used by the agent to draft and apply changes to the file system.
  • Ingestion points: Markdown files located in the thoughts/ directory or other project files containing review markers.
  • Boundary markers: The skill uses HTML comment tags (<!-- review-start -->, <!-- review-end -->) as delimiters.
  • Capability inventory: The agent has the ability to execute shell commands (via the file-review binary and bun) and modify local files.
  • Sanitization: The workflow incorporates a mandatory human-in-the-loop step via AskUserQuestion, requiring the user to approve edits before they are applied.
  • [COMMAND_EXECUTION]: The skill executes local binaries and manages complex installation processes.
  • Evidence: Instructions include running brew install, bun install, and launching the file-review GUI tool.
  • Tunneling Feature: The skill supports a --tunnel flag which uses localtunnel to expose a local web server to the internet for remote review sessions.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of software from external sources.
  • Evidence: The skill fetches source code from https://github.com/desplega-ai/ai-toolbox.git and uses Homebrew taps to install the tool. These resources originate from the skill author's infrastructure.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 8, 2026, 11:34 PM