delegate-htmldoc
Fail
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: HIGHCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill's core functionality is delegated to a JavaScript file named
delegate-cli.mjs. This file is invoked by almost every script in the package (e.g.,run.sh,prepare.sh,dispatch.sh,htmldoc-permalink.sh) usingnode. However, the source code fordelegate-cli.mjsis not provided in the skill files, representing an unverifiable execution of bundled code. - [COMMAND_EXECUTION]: The skill uses
Bashto execute multiple internal scripts that wrap the Node.js execution. While the wrappers themselves are simple, they pass all command-line arguments directly to the hiddendelegate-cli.mjstool, which performs complex tasks like dispatching subagents and resolving models. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (Markdown requests) to generate HTML documents. While
references/styleguide.mdprohibits the inclusion of JavaScript in generated content, the toolscripts/htmldoc-permalink.shis explicitly designed to perform content injection into the generated HTML files. This creates a surface where malicious instructions could be embedded or manipulated. - Ingestion points:
stdinforscripts/run.shandscripts/prepare.sh. - Boundary markers: The skill uses Markdown section headers (Objective, Scope, Context) to delimit input.
- Capability inventory: The skill possesses extensive capabilities including
Bashscript execution, file reading (Read), and user interaction (AskUserQuestion). - Sanitization: The skill relies on natural language instructions to prevent JavaScript generation by subagents, but provides an automated tool for post-generation injection.
Recommendations
- AI detected serious security threats
Audit Metadata