Read-only OpenAI safety workflow. It counts Responses API input tokens and runs moderation on explicit caller-provided text or image inputs, then summarizes safety categories and readiness signals. It never creates model output, enables tools, uploads/downloads files, mutates vector stores, uses realtime, or calls arbitrary OpenAI routes.

How it works 7 steps

Inspect the data fetches, transforms, gates, and output this recipe runs.

Process (7 steps)

■ input_tokens ?

■ moderation ?

■ moderation_results default

Apply default

■ result_count count

Count items in moderation results

■ summarize ?

■ moderation_table to_table

Format results as a data table

■ card to_summary

Format results as a summary card

Settings 7 configurable

Configurable at install. Defaults shown — change them anytime in Recued.

focus setting = Summarize moderation flags, category scores, applied input types, input-token count, and practical next steps. Do not invent facts outside the OpenAI moderation or token-count responses.

openai setting =

input text setting =

max length setting = 900

instructions setting = Evaluate this input for a governed assistant workflow.

response model setting = gpt-5.5

moderation model setting = omni-moderation-latest