The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection through its handling of token data.
Ingestion points: The skill ingests data from external token files (*.tokens.json) through the dtokens/model.py module.
Boundary markers: The system lacks delimiters to isolate ingested token data from internal instructional templates.
Capability inventory: The skill can write files, generate shell commands, and create HTML content with external dependencies.
Sanitization: Token names and values are interpolated into generated prompt strings in dtokens/export_prompt.py without sanitization or escaping, which could allow malicious token data to influence downstream AI models.
[COMMAND_EXECUTION]: The skill involves the generation of CLI commands and the execution of a local utility server.
The prompt command generates ready-to-paste shell command strings for external scripts (gpt_image_2.py and nano_banana.py) which are not included in the package.
The serve and use commands launch a local HTTP server using Python's http.server module, binding to the loopback address (127.0.0.1), and automatically open the system's web browser.
[EXTERNAL_DOWNLOADS]: The skill and its included examples reference various external assets and developer tools from well-known services.
HTML previews and landing page examples reference resources from Google Fonts, JSdelivr, Cloudflare, UNPKG, and ESM.sh.
These references are standard for the skill's purpose and utilize reputable content delivery networks.

design-tokens