Idea intake

Idea Intake

Required Knowledge

• docs/film-preproduction/chat-co-creation-interface.md
• docs/film-preproduction/creative-copy-deck-capability.md
• docs/film-preproduction/schemas/project.yaml
• docs/film-preproduction/research/channel-playbooks.md

Inputs

• raw user idea
• optional channel, duration, aspect ratio, references, constraints

Outputs

• project brief
• parsed intent
• channel inference
• facts, source evidence, and labeled inferences
• audience tension, business objective, single-minded proposition, and avoid-cliches
• missing decisions
• recommended next skill: director-room

Chat Surface

Show a brief in chat before writing downstream artifacts:

• 阶段: 想法读取
• 智能体创作内容: restated idea, inferred channel, likely story duration, generation-unit cap, aspect, constraints, known facts and evidence, explicit inferences, audience tension, business objective, single-minded proposition, and avoid-cliches.
• 用户确认点: ask whether the brief is right before moving into director-room options.

Ask one question only: "这个 brief 是否正确？"

Creative Foundation Gate

Before recommending a concept:

• Record each factual input under creative_foundation.facts and bind it to one or more source_evidence_ids.
• Record source type, locator, summary, and verification status under creative_foundation.source_evidence.
• Keep DIRcreative interpretations under creative_foundation.inferences; name the supporting fact IDs and confidence.
• Treat evidence, facts, and inference IDs as fail-closed: arrays only, mapping items only, unique non-empty IDs, and resolving references.
• For commercial work, fill commercial_strategy.audience_tension, business_objective, single_minded_proposition, and avoid_cliches.
• Resolve audience-tension evidence IDs and proposition support-fact IDs before the brief can pass.
• Treat generic goals such as "build awareness" or "make it cinematic" as unresolved until the audience pressure and intended change are concrete.
• Do not convert an unverified input or an inference into an audience-facing factual claim.

If source evidence or audience tension is missing, keep the brief at needs_user and name only the smallest missing decision. Do not compensate with invented research or polished copy.

Rules

• Infer a channel only when evidence is strong.
• Keep uncertainty explicit.
• Distinguish story_duration from generation_unit_limit. A 15s video model cap means each generated group is at most 15s, not necessarily that the story is 15s.
• If the duration meaning is ambiguous, ask one clarification question before scripting.
• Separate facts from inferences in both the artifact and the chat preview.
• Do not write a single-minded proposition that has no supporting fact IDs.
• Record category cliches and familiar visual shortcuts the downstream concept should avoid.
• Do not write story, script, shot list, image prompts, or video prompts.

skill_run_receipt

Include skill_run_receipt with input artifacts, output artifacts, fail-closed ID status, source-evidence and inference binding status, tension/SMP source status, commercial-strategy status, decisions, unresolved questions, QA status, and next_recommended_skill: director-room.