aside-browser

Fail

Audited by Gen Agent Trust Hub on Oct 22, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The installation instructions in README.md and docs/index.html advise users to execute a remote script via curl -fsSL https://releases.aside.com/install.sh | bash. This pattern is a high-risk operation as it executes unverified remote code directly in the user's shell.
  • [COMMAND_EXECUTION]: The skill relies on executing the aside CLI tool (aside exec and aside repl) to function. These commands spawn sub-agent sessions and interactive JavaScript REPLs that can execute further system-level operations.
  • [DYNAMIC_EXECUTION]: The aside repl feature provides a persistent JavaScript environment (ES2023+) where the agent can execute arbitrary code. This environment includes access to powerful globals like fs (file system), Buffer, and fetch (network), enabling complex runtime execution paths.
  • [DATA_EXFILTRATION]: By design, the skill provides the agent with access to the user's cookies, logged-in accounts (e.g., Slack, LinkedIn), and browsing history. While SKILL.md contains warnings against cross-origin credential forwarding, the technical capability to read sensitive data and send it externally via fetch() or page.pdf() exists.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it is designed to ingest and process untrusted data from external websites.
  • Ingestion points: The snapshot() and annotatedScreenshot() functions in SKILL.md are used to read webpage content and accessibility trees into the agent's context.
  • Boundary markers: The skill uses virtual ref IDs (e.g., e31) to identify elements, which provides some structural separation, but does not strictly prevent the agent from obeying instructions embedded in the webpage text.
  • Capability inventory: The agent has the ability to write files (fs.writeFile), perform network requests (fetch), and execute browser actions (click, openTab).
  • Sanitization: The instructions recommend using snapshot() over page.content() to minimize exposure to raw HTML, and advise against using fetch() for cross-origin credential forwarding.
Recommendations
  • HIGH: Downloads and executes remote code from: https://releases.aside.com/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 22, 2026, 12:17 PM