The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its tool-calling capabilities.
Ingestion points: User prompts are ingested in backend templates such as templates/python-backend/fastapi-chat.py and templates/nextjs/api-chat-route.ts.
Boundary markers: While system prompts are used (e.g., in templates/nextjs/api-chat-route.ts), they do not explicitly instruct the model to ignore instructions embedded within data fetched by tools.
Capability inventory: The skill defines several tools in templates/shared/tool-schemas.ts with significant capabilities, including web_search for fetching external data and send_email for outgoing communication.
Sanitization: The skill utilizes Zod for schema validation (e.g., webSearchSchema), ensuring structured data but not preventing natural language instruction injection within those fields.
[EXTERNAL_DOWNLOADS]: The skill includes dependency configuration files (package.json, requirements.txt) and installation scripts (scripts/install-dependencies.sh) that fetch established packages from official registries (NPM, PyPI). These include libraries such as openai, fastapi, and the thesys-genui-sdk.
[COMMAND_EXECUTION]: The scripts/check-versions.sh script executes a local node command to verify installed package versions, which is a standard development utility behavior.

thesys-generative-ui