v3.7.0

How to install

Option 1: the skill (any assistant):

npx skills add Freakling/Godot-Director

In Claude Code this becomes a /godot-director command. The skill fetches Godot Director and runs onboarding.

Option 2: the Claude Code plugin:

/plugin marketplace add Freakling/Godot-Director
/plugin install godot-director@godot-director

Open a new session in your game, then run /godot-director:godot-director.

Option 3: manual:
Put this repository next to project.godot as a folder named Godot-Director, then ask your assistant to read Godot-Director/ONBOARDING.md and follow it.

Upgrade:

• Skill: ask for the skill again (/godot-director in Claude Code); it fetches the latest Godot Director each time.
• Plugin: run /plugin marketplace update godot-director and then /plugin update godot-director@godot-director. Start a new session in the game, and run /godot-director:godot-director again.
• Manual: put the new Godot Director folder in the game, and ask for ONBOARDING.md again.

---

Designed to work with any AI coding assistant that reads AGENTS.md. Built and tested on Claude Code; other assistants are untested.

---

What's changed

• New: /research, the research assistant (.godot-director/procedures/research.md). When a design question depends on facts nobody has checked, or you ask for research, a read-only researcher subagent looks into it in a fresh context and reports a note. It informs; you decide: its recommendation is one option shown with its reason, and it never records a decision, edits the GDD or design/decisions.md, answers an open question, or asks you anything. It states its assumptions in the note instead.
  • Research is a task type. Research items get their own IDs, R<n>, drawn from the Next IDs line like bugs. Instead of Done when they have Question: and, when they feed an open question, Informs: Q<n>. They are sized like any item, and the size picks the model. An item is done when its note exists.
  • Two ways to start. On the fly: a design session delegates a complex question, or you run /research <question>, and the item is created and run at once. For later: queue the item, and list it under Depends on of the tasks that need its findings. A task that needs the decision stays gated by the open question itself.
  • /next-task hands research items to the researcher. No check runs and no review: research changes no code. This session writes the note to design/research/YYYY-MM-DD-<slug>.md and commits it with the records. A finished item that informs Q<n> makes that question ready for a design call, and a run goes on to the next item.
  • Design sessions list design/research/ first, show notes on the topic in the context brief, delegate a question that needs facts, and end the decision's why with the note's path so later sessions reuse the evidence.
  • The note is one dated file per topic: the conclusion first, the options and trade-offs, the recommendation, every claim with its source, source tier and confidence, the date the claims were verified, and what stays unknown.
  • A Godot domain file (.godot-director/research-domain.md) is read when the question touches the engine and its docs for the project's version, plugins and assets and their licences, export and store requirements, or how comparable games solve the problem (design references, checked against the Design Pillars).
  • Confidentiality. Project text and unpublished design never go into a public search query or fetch.
  • The researcher has Read, Grep, Glob, WebSearch and WebFetch: no edit, write or shell tool. .claude/settings.json allows web search and fetch; everything with side effects keeps asking.
• Align checks that Next IDs has an R counter, that research items have a Question: instead of Done when, and that each done research item has its note.
• Selftest checks that the research files exist and are wired into the procedures, the item format, the seeds and the docs, that the researcher has no edit, write or shell tool, that the installer ships everything (including with --tools none), and that a research note commits without running the check.
• The example (examples/market-day) shows the R counter and the note folder, and the scenarios gain two for research.
• Web search and fetch are allowed in every session in a game, not only for the researcher, because Claude Code can't limit a permission to one subagent: .claude/settings.json allows WebSearch and WebFetch, so any session can search and fetch the web without asking.

Upgrade steps

1. TASKS.md: add an R counter to the Next IDs line, so **Next IDs:** T12 · B3 becomes **Next IDs:** T12 · B3 · R1.
2. design/research/: the installer creates the folder with a short README.md (a project file, yours from then on) when the game doesn't have it. Notes are added there as research runs; commit them like the other records.
3. .claude/settings.json is framework-owned, so it is replaced on upgrade. If you edited it, the new version arrives as .claude/settings.json.gdir-new; merge it, and make sure "WebSearch" and "WebFetch" are in permissions.allow, which lets the researcher read the web without asking each time. Restart Claude Code afterwards so the new skill, subagent and permissions load.

No other game-owned file changes are required.