The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses eval() and new Function() in adventure_runtime.py, adventure.js, and engine.js to compile and execute logic such as exit guards, effects, and scores directly from adventure JSON/YAML data. This architecture enables arbitrary code execution if the source data is maliciously crafted.
[DATA_EXFILTRATION]: The GitHub OAuth implementation in github-api.js uses window.opener.postMessage('*') to communicate authorization codes. Using a wildcard origin allows any website that opens the window to potentially intercept these sensitive tokens.
[CREDENTIALS_UNSAFE]: Multiple components, including api-keys.js, github-api.js, image-generate.js, and image-analyze.js, store API keys for services like OpenAI, Anthropic, Google, and GitHub in the browser's localStorage in plaintext. This makes the keys vulnerable to exfiltration through any Cross-Site Scripting (XSS) vulnerability on the site.
[COMMAND_EXECUTION]: The merge command in adventure.py calculates file paths for state updates using input from a JSON file (op['target']) without sanitization. This presents a potential path traversal vulnerability where a malicious state file could overwrite files outside the intended adventure directory.
[EXTERNAL_DOWNLOADS]: The skill frequently interacts with external APIs from OpenAI, Anthropic, Google, Stability AI, Replicate, and GitHub to generate images, analyze content, and manage source code. While these are well-known services, the skill's operation depends on external infrastructure and user-provided credentials.

adventure