The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The file references/capacitor-ios.md contains explicit instructions to execute commands with sudo (e.g., sudo gem install cocoapods) to resolve environmental build errors, which bypasses standard permission constraints.
[DYNAMIC_EXECUTION]: The script scripts/with_server.py utilizes subprocess.run() with the shell=True parameter to execute command strings passed as CLI arguments. This pattern facilitates the execution of arbitrary shell commands and increases the risk of command injection if the input strings are not strictly controlled.
[COMMAND_EXECUTION]: The skill makes extensive use of shell command execution across various phases, including execSync calls in documentation examples and direct script invocations for testing (with_server.py) and image comparison (imgdiff.py).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data sources, creating a significant injection surface:
Ingestion points: The agent reads project-level files such as package.json, assets_index.json, and directory structures to determine its workflow.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard instructions that might be embedded within these processed files.
Capability inventory: The skill has broad capabilities including file writing, shell execution via the Bash tool, and project deployment to remote hosts.
Sanitization: No validation or sanitization is performed on the contents of ingested files before they are used to influence the agent's logic or command arguments.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of numerous third-party packages from registries like npm, PyPI (Pillow), and RubyGems. These are documented as standard project dependencies.
[REMOTE_CODE_EXECUTION]: The references/game-audio.md file includes instructions to import executable code from the unpkg.com CDN at runtime using ES module imports (import { repl } from 'https://unpkg.com/...').

game-pipeline