manage-apps-and-sounds-headless

Warn

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides explicit instructions to bypass environmental security controls, such as sandbox proxies, by using env -u HTTPS_PROXY -u HTTP_PROXY and curl --noproxy '*'. This behavior can be used to evade network monitoring and logging.
  • [COMMAND_EXECUTION]: The skill executes multiple local shell scripts and binaries to perform its tasks.
  • resolve_pushover_secret.sh is used to load sensitive credentials into environment variables.
  • find_jingles.sh and make_custom_sound.sh are executed for media processing.
  • bun is used to execute the core logic in pushover_headless_web_control.ts.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs external binaries and media content.
  • It uses bunx playwright-core install chromium to download the Chromium browser for testing.
  • It references sourcing MP3 files from Mixkit, an external third-party site.
  • [DYNAMIC_EXECUTION]: The skill implements a 'Self-Evolving' pattern where it instructs the AI agent to 'fix this file immediately' if issues are encountered. This encourages the agent to autonomously modify its own instruction set (SKILL.md) at runtime, which is a form of self-modifying behavior.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the web.
  • Ingestion points: It scrapes content from the pushover.net dashboard (application names, descriptions, sound lists) and processes external MP3 metadata.
  • Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands in the data it processes.
  • Capability inventory: The skill has the ability to execute shell commands, bypass proxies, and rewrite its own source instructions.
  • Sanitization: There is no evidence of sanitization for the data scraped from the web before it is used in logic or written to the skill file.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 7, 2026, 02:12 AM