The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bash command block to perform a syntax check on generated JavaScript workflow files. It employs node -e to read the script, wrap it in an asynchronous function block, and write it to a temporary location (/tmp/wf.mjs). It then uses node --check to verify the code's syntax without executing it.
[PROMPT_INJECTION]: Native workflows utilize an agent() hook that accepts a prompt string, which constitutes an ingestion point for indirect prompt injection if external data is subsequently interpolated.
Ingestion points: The prompt argument within the agent() function in .claude/workflows/*.js files.
Boundary markers: No specific delimiters or instructions to ignore embedded instructions are included in the workflow templates.
Capability inventory: Workflow scripts can utilize all available agent tools, including file system access (Read, Write, Edit) and shell execution (Bash).
Sanitization: The instructions do not define specific sanitization or escaping protocols for data passed into the agent hook.
[SAFE]: All file operations and script generation are restricted to the local project environment (.claude/workflows/) and are consistent with the skill's primary purpose of creating orchestration logic.

workflow-create