DaVinci Resolve Reference Skill

A reference skill for Agent Kaizen covering the DaVinci Resolve Reference Manual (the Blackmagic Design editing, color, VFX, and audio app). The prose is original (recontextualized); identifiers — page/panel/node names, control and parameter labels, menu paths, keyboard shortcuts, enum values, numbers, and units — are preserved exactly.

What this skill covers

This skill is a faithful reference for DaVinci Resolve. It covers the Media, Cut, Edit, Fusion, Color, Fairlight, and Deliver pages; importing and managing media, proxies, and the Media Pool; timelines, editing tools, trimming, transitions, retiming, and keyframing in the Inspector; color grading (primaries, curves, qualifiers, power windows, tracking, nodes, scopes, color management RCM/ACES, LUTs); Resolve FX and OpenFX; Fusion compositing and the node editor; Fairlight audio editing, mixing, and effects; rendering and export on the Deliver page; and project/database management, collaboration, and Blackmagic Cloud. It covers the DaVinci Resolve 20.3 Reference Manual plus the DaVinci Resolve 21 New Features Guide. It is a flat skill: a single SKILL.md plus a references/ library of one focused topic file per subject. It is not for other NLEs (Premiere, Final Cut, Avid).

What's inside

• SKILL.md — the trigger description, when-to-use, workflow, and routing into the references library.
• references/ — right-sized topic files (one subject per file) plus INDEX.md (start here) and topics.json (metadata).
• GOTCHA.md — recurring failure modes and what to do instead.

Status

Recontextualized into original prose (identifiers preserved); packaged in the skills store and ready to publish as the public AI-SKILL-davinci-resolve repository. Publication is performed by the maintainer.

Reducing idle context cost (skill invocation policy)

Every installed skill costs a little context on every session: the agent sees each skill's name and description before you ever use it. If you rarely use this skill in a given project, you can keep it installed and still explicitly invocable while hiding it from the model's automatic listing.

Doing so does not modify this skill's source repo — the policy lives in your local agent settings. The Agent Kaizen skill-drafting repo ships a manager, scripts/skill_policy.py, that sets this for all your skills at once:

python skill_policy.py audit     # list every skill + its current policy + idle cost
python skill_policy.py plan      # write a decision file with recommendations (nothing applied)
# edit that decision file: set selected_policy + approved:true for the skills you choose
python skill_policy.py preview   # show the exact change
python skill_policy.py apply     # apply ONLY what you approved (backup + rollback recorded)
python skill_policy.py restore   # roll back

• Claude Code (works today): writes skillOverrides: { "<skill>": "user-invocable-only" } to .claude/settings.local.json — zero idle listing cost, still available from the /skills menu. Start a new session for it to take effect; invoke it any time via /skills.
• Codex (currently unreliable): explicit-only Codex skills are affected by an open bug (openai/codex#23454) where $skill invocation of an explicit-only local skill can fail. Until it's fixed, leave Codex skills implicit, or fully disable rarely-used ones with [[skills.config]] enabled = false in config.toml. The manager audits Codex but does not change Codex policy.