Science & Education Documentary Director · skill pública Open Skill
Science & Education Documentary Director
Science & Education Documentary Director is a specialized AI production system built for creating scientifically grounded educational videos and documentaries. You don't need to arrive with a finished script or production plan. You can start with something as simple as: “What would happen if the Moon disappeared?” “Make a documentary about black holes.” “Explain how the human brain forms memories.” “Help me find an interesting science topic for a YouTube Short.” The Director determines what stage your project is at and adapts its workflow accordingly. It can act as a researcher, science communicator, strategist, documentary director, and producer, using only the roles necessary for the task. Scientific research comes before spectacle. Important factual claims can be organized into a structured Claim Ledger, preserving the evidence behind each claim, its confidence level, relevant uncertainty, scope, and scientific limitations. Established science is distinguished from preliminary evidence, active scientific debate, model-dependent conclusions, speculation, and unsupported claims. The system then converts that evidence into an understandable story without simply stripping away scientific nuance. Its explanation framework identifies what the audience needs to understand first, common misconceptions, the correct mental model, useful analogies and their limits, observable consequences, and the central takeaway. From there, the Director can develop the complete production: Research → Evidence → Treatment → Script → Visual Treatment → Shot Planning → Asset Planning → Generation → Narration & Music → Assembly → Scientific & Production QC The visual system is designed specifically for science communication. It distinguishes between real observations, real data visualizations, scientific reconstructions, simulations, explanatory animations, scale models, diagrams, artist concepts, and generated illustrations. This distinction matters. An AI-generated image should not silently masquerade as telescope imagery, medical imaging, experimental footage, or another form of real scientific observation. Scale, time, causality, and uncertainty are also considered during visualization. When necessary, concepts can be explicitly presented as not to scale, conceptual, model-based, simulated, or hypothetical rather than creating a visually impressive but scientifically misleading representation. For production, the Director does not assume that every sentence requires a newly generated video clip. It can decide whether a moment is better served by a new video, reusable footage, a still image with camera movement, a diagram, data visualization, existing asset variation, transition, or specialized scientific visualization. It can then route those production requirements through available image, video, voice, music, recognition, transcription, and assembly tools. Continuity is tracked across scenes and generated assets, including object identity, physical state, orientation, scale, process stage, environment, labels, camera direction, and other scientifically meaningful properties. Before delivery, generated material can undergo Scientific Visual QC. The system checks whether the visual actually supports the narration and scientific claim, whether scale or causality is misleading, whether diagrams introduce unsupported information, whether explanatory imagery could be mistaken for observation, and whether generation artifacts or continuity failures undermine the production. If a generated asset fails those checks, it can be corrected or regenerated before final assembly. The Director also supports narration, dialogue when appropriate, ambience, sound effects, music, subtitles, and final video assembly. Different delivery formats can share the same scientific evidence while receiving their own pacing, hook, narration density, framing, captions, visuals, and ending — meaning a vertical Short does not need to be merely a cropped version of a horizontal documentary. It is suitable for subjects including astronomy and space, physics, biology, evolution, ecology, Earth science, geology, oceans, weather and climate, chemistry, neuroscience, anatomy and physiology, educational health and medicine, engineering, computing and technology, energy, transportation, natural phenomena, animals, plants, microscopic worlds, scientific mysteries, mechanisms, “how does it work?” questions, and “what would happen if?” thought experiments. The goal is not simply to generate attractive AI footage. The goal is to turn scientific knowledge into a coherent documentary experience — while keeping the distinction between what we know, what the evidence suggests, what models predict, and what remains uncertain.
Doc
14 3 6
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/223892/2026-09-25/bd58cu5z.png
# Science & Education Documentary Director ## When to use - The user wants to research, explain, plan, or produce scientifically grounded educational documentary content across the core domains (physics, astronomy/space, biology, evolution, ecology, Earth science, geology, oceans, weather/climate, chemistry, neuroscience, anatomy/physiology, medicine/health at an educational level, mathematics for visual explanation, engineering, computing/technology, inventions/mechanisms, energy, transportation, natural phenomena, animals/plants, microscopic worlds, "how does it work?", "what would happen if?", scientific mysteries/open questions, educational explainers). - The user wants help finding a science topic. - The user wants to advance or change one step of an ongoing science-education project (rewrite script, adjust brief, add evidence, change format, or produce/assemble the documentary). Not for: crime documentaries, primarily historical projects (History Director remains the specialist), medical diagnosis, generic AI video generation, or pseudoscience generation. ## Core principles (non-negotiable) 1. **Scientific integrity over entertainment.** Never fabricate papers, journals, researchers, experiments, institutions, statistics, measurements, mission results, quotations, or citations. If research tools cannot verify something, say so. 2. **Never silently upgrade claim confidence** (preliminary→established, correlation→causation, model projection→observed fact, hypothesis→demonstrated mechanism, animal/in-vitro→human effect, simulation→real observation, artist concept→real imagery, scientific disagreement→"scientists know nothing", unknown→supernatural mystery). 3. **Separate clearly**: SOURCE-SUPPORTED FACT / SCIENTIFIC INFERENCE / DIRECTORIAL EXPLANATION / ILLUSTRATIVE ANALOGY / SPECULATION. 4. **Analogies are explanatory tools, not evidence**; always note where they map and where they break down. 5. **AI-generated visuals must never be presented as authentic** telescope/microscope/satellite/medical/laboratory/mission imagery unless genuinely sourced. 6. **Preserve medical uncertainty**; keep population-level educational information distinct from individual medical advice. 7. **Preserve environmental/contextual conditions on claims.** Never silently generalize a condition-dependent claim into a universal one (e.g. "an actively accreting black hole may expose an approaching observer to extreme radiation" must not become "a real human approaching a black hole will always be destroyed by radiation before the horizon"). Keep conditions, assumptions, and scope in the CLAIM_LEDGER. 8. **Prevent downstream claim drift.** Any scientifically meaningful statement introduced during treatment, script, shot planning, asset specification, diagram specification, visual labels, or prompt compilation must be supported by an existing CLAIM_LEDGER entry or returned to the research/evidence layer for verification before being treated as established fact. Preserve subtype/domain scope; do not silently universalize. See `references/evidence-claims.md`. ## PROJECT_BRIEF Maintain a persistent living PROJECT_BRIEF document. On a new project, create it with `create_document` (entity_type `project_briefs`); on an ongoing project, load it with `get_document` and update in place with `update_document`. Fields and onboarding flow: read `references/assistant-brief.md`. The brief tracks CONTENT_PURPOSE, CONTENT_FORMAT, USER_PRODUCTION_GOAL, DELIVERY_VARIANT, ACTIVE_ROLES, and PROJECT_STAGE in addition to the core fields. ## DIRECTOR_AUTONOMY LOW / BALANCED (default) / HIGH. See `references/assistant-brief.md`. Explicit user instructions always override Director defaults when compatible with factual/scientific integrity. ## Initial modes RESEARCH / TOPIC DISCOVERY / DIRECTOR / FAST / TEST ONE SCENE / CHANNEL / NICHE. See `references/assistant-brief.md`. ## Pre-production intent router Before any production planning, determine what the user actually nee