DirectorMind is an evidence-led research repository for turning a locked literary scene into model-independent directing decisions: shot boundaries, POV, coverage, blocking, reaction priority, pacing, sound, continuity, and AI-safe fallbacks.
It is not a collection of film clips, a director-name imitation pack, or a provider-specific prompt library. Real scenes are studied as evidence; transferable rules remain candidates until their triggers, boundaries, counterexamples, and forward tests are documented.
- Read
context/00_START_HERE.md. - Read
context/STATE.md. - Use
skills/drama-director-compiler/SKILL.mdwhen compiling an approved script. - A second Codex instance should start with
prompts/ROOT_AGENT_START_PROMPT.md.
- model-neutral Director IR contract and JSON Schema;
- deterministic validator and Markdown renderer;
- five general directing rules plus seven Good Omens-derived transfer candidates;
- a ten-series short-drama mechanism snapshot used only to seed scene-problem research;
- private-project forward-test lessons, without publishing the underlying scripts or episode IR.
The baseline is a research starting point, not a universal directing model and not production authorization.