The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes workspace_execute_command to perform code validation using a utility located at node_modules/@n8n/workflow-sdk/dist/cli/index.js. This execution is scoped to the workspace and serves to verify the syntax of generated TypeScript code.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by generating code based on user-provided instructions. This risk is managed by a mandatory validation process and explicit guidance on data handling.
Ingestion points: User prompts provided to the agent and source files fetched from the n8n workspace (SKILL.md, Step 5).
Boundary markers: The agent is instructed to use placeholder() for sensitive or undiscoverable values, preventing hardcoding (SKILL.md, Placeholders section).
Capability inventory: The skill has the ability to write files (workspace_write_file), edit files (workspace_str_replace_file), and execute validation commands (workspace_execute_command).
Sanitization: Mandatory use of workflow-sdk validate to check the generated source file for errors and warnings before deployment (SKILL.md, Step 7).
[PROMPT_INJECTION]: The skill contains an instruction for the agent to suppress intermediate output ('Do not produce visible output until the final step, unless blocked.'). In the context of complex automated workflow generation, this is a standard operational guideline to improve user experience rather than a malicious attempt at concealment.

workflow-builder