!
Purpose & Capability
The skill claims only to map user intent to a 'node zuma.js' invocation, but the included code does much more: it reads a .env and process.env variables (ZUMA_SERVER_URL, API_KEY, TOKEN, IMGBB_API_KEY, DOWNLOAD_LINKS, etc.), performs HTTP requests to local and remote endpoints, synchronizes guide.md into the skill directory, and references remote download URLs for a Zuma desktop executable. Many of these capabilities (network downloads/uploads, registry access) are beyond a simple command-mapper and are not justified in the description.
!
Instruction Scope
SKILL.md's runtime instructions insist the agent only run node zuma.js commands and forbid using system shell commands, but the implementation performs actions that go beyond 'just run and return output': it may trigger npm installs, call local and remote HTTP endpoints, copy files into the skill directory, and the script uses child_process.execSync (registry queries). The skill documentation does not declare or explain these reads/writes, nor the use of external image upload services.
!
Install Mechanism
There is no install spec (instruction-only), but the code expects to run 'npm install' or 'pnpm install' when dependencies are missing. package.json pulls in node-fetch and form-data (reasonable), however the code also contains arrays of remote download URLs (GitHub/Gitee .zip releases) for a Windows executable — downloading/extracting those at runtime would be high risk. The repository also contains an embedded default IMGBB API key. No explicit, vetted release hosts or checksums are provided.
!
Credentials
The skill declares no required environment variables, yet zuma.js reads many env vars and a .env file (ZUMA_SERVER_URL, API_KEY, TOKEN, IMGBB_API_KEY, DOWNLOAD_LINKS, etc.). A non-trivial secret (IMGBB_API_KEY) is hard-coded in defaults. The mismatch between declared requirements (none) and actual env/credential usage is a clear proportionality problem and increases exfiltration risk (screenshots/images may be uploaded).
ℹ
Persistence & Privilege
The skill does not set always:true and is user-invocable. The code writes files (syncGuide copies guide.md into the skill root), may create workspace directories under the user's home, and reads the Windows registry to find an installed exe path. Those behaviors are plausible for a desktop agent but should be considered persistent and able to modify files under user home. No evidence it modifies other skills or system-wide agent settings.
Scan Findings in Context
[hardcoded-imggb-api-key] unexpected: zuma.js contains a default IMGBB_API_KEY value in source. Hard-coded API keys in code are unexpected and increase risk (exfiltration, misuse).
[child_process_execSync] expected: The code uses child_process.execSync (e.g., to query HKCU registry for InstallPath). Registry lookup is plausible for a Windows desktop agent to find installed components, but it contradicts the SKILL.md prohibition on system commands at the agent instruction level.
[external-network-calls] expected: The implementation makes HTTP requests to a local Zuma Robot server (127.0.0.1:53030) and can upload images to api.imgbb.com and to a local proxy (localhost:19000). Network calls are expected for an agent that controls a desktop service and uploads screenshots, but users should be aware uploads go to external hosts (imgbb) and remote download links are listed.
[reads-undisclosed-environment-variables] unexpected: zuma.js reads many environment variables and a .env file but the skill metadata declared none. This mismatch is unexpected and relevant to trust decisions.
What to consider before installing
This skill contains executable code that does more than 'map a command' — it reads .env and process.env variables, queries the Windows registry, writes/copies files into the skill folder, may download remote release archives, and uploads images to an external image host (imgbb) using a hard-coded API key. Before installing: 1) only install if you fully trust the skill's source; 2) review the full zuma.js and upload.js code yourself (or have someone you trust do it); 3) remove or replace the hard-coded API key and explicitly set any required environment variables rather than relying on defaults; 4) run the skill in a restricted sandbox or VM and monitor network activity (especially outbound uploads and any downloads); 5) if you need only screenshot-to-local functionality, consider modifying the code to disable external uploads; 6) if anything is unclear, treat this skill as potentially data-leaking and avoid granting it access to sensitive accounts or files.
