financeflash/unvibecode/llm-content-generation-and-output-validation

Write a LLM Structured Content Generation Specification

Use application-native terms. Decide:

• Who may generate which content using which tenant data
• Permitted input, context, attachment, personal-data, and secret boundaries
• System-instruction and user-input separation and precedence
• Prompt template, model, model version, parameters, and rollout policy
• Token, character, attachment, latency, cost, and concurrency limits
• Required output schema, field constraints, enums, formats, and unknown-field policy
• Maximum parse, validation, repair, and provider retry attempts
• Safety, moderation, refusal, human-review, and fallback policy
• Whether and how raw prompts, responses, usage, and accepted outputs are retained
• Caching, idempotency, observability, privacy, and abuse policy

Write scope, actors, objects, states, paths, user outcomes, permissions, validation, recovery, privacy, misuse, acceptance criteria, and unanswered decisions.