ipollowork-template-generation
Warn
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
SKILL.mdinstructions direct the agent to execute a shell command:node .codex/skills/ipollowork-template-generation/scripts/audit-template.mjs <template-directory>. This allows the agent to trigger local script execution with variable user input. - [REMOTE_CODE_EXECUTION]: The script
scripts/audit-template.mjsconstructs a JavaScript string (validationProgram) and executes it usingbun -e. This string includes animportstatement for a module path (validatorUrl) that is dynamically resolved at runtime based on the repository's file structure. This evaluator pattern for running dynamically generated code is a risk vector for execution control. - [PROMPT_INJECTION]: The skill processes user-supplied template directories containing HTML, CSS, and JSON files. 1. Ingestion points: The
<template-directory>read inSKILL.md. 2. Boundary markers: None specified in the instructions to prevent the agent from following instructions embedded in the data. 3. Capability inventory: Shell command execution (node,bun) and file system access. 4. Sanitization: The script usesJSON.stringifyfor internal variables, but the agent lacks explicit sanitization instructions for the ingested content. This surface facilitates indirect prompt injection risks.
Audit Metadata