The Agent Skills Directory

[PROMPT_INJECTION]: The template BuildIntentChain.ts is susceptible to indirect prompt injection because user-provided text is interpolated directly into a system prompt without the use of delimiters or escaping mechanisms.
Ingestion points: The RunWorkflowUseCase.ts accepts a userInput string from the RunWorkflowDTO which is passed to the intent classifier.
Boundary markers: None are present in the BuildIntentChain.ts prompt template.
Capability inventory: The skill includes scripts/copy_templates.sh which can perform directory creation and file system writes.
Sanitization: The skill mitigates risks through strict validation in ConversationIntent.ts, which uses an allow-list enum and confidence thresholds for all LLM-derived data.
[COMMAND_EXECUTION]: The skill includes a shell script scripts/copy_templates.sh that executes mkdir -p and cp -R commands using a user-provided target directory argument to scaffold the project structure.
[EXTERNAL_DOWNLOADS]: The templates provided in the skill reference several external Node.js packages that would be required for the generated code to function, including @langchain/langgraph, @langchain/core, @langchain/openai, and @andireuter/js-domain-principles.

ddd-langchain-langgraph