Tooling

.claude/commands/vt-diff.md

@@ -0,0 +1,84 @@
+---
+description: Diff Vanilla Tweaks against our packs.json, research Bedrock feasibility, and prepare the tracking issues
+allowed-tools: Bash, Read, Edit, Write, Glob, Grep, WebFetch, Agent
+---
+
+Run the Vanilla Tweaks parity check.
+
+**Read `tools/vt-diff/AGENTS.md` before doing anything else.** It holds the rules, the source
+hierarchy and the verdict rubric. This command is the running order; that file is the method.
+
+Two rules override everything below: never run `issues.mjs --apply`, and never record an alias or a
+verdict you have not verified.
+
+## 1. Confirm the pinned Vanilla Tweaks version
+
+Read `vanillaTweaks.version` from `tools/vt-diff/config.json`, then probe the next few candidates —
+Vanilla Tweaks uses `YY.N` keys (`26.1`, `26.2`, ...):
+
+```sh
+for v in <pinned> <next> ; do curl -s -o /dev/null -w "$v=%{http_code}\n" \
+  "https://vanillatweaks.net/assets/resources/json/$v/rpcategories.json"; done
+```
+
+If a newer key returns `200`, update `config.json` and say so in your summary. Do not bump silently.
+
+## 2. Fetch and diff
+
+```sh
+node tools/vt-diff/fetch.mjs && node tools/vt-diff/diff.mjs
+```
+
+## 3. Alias pass
+
+Follow `AGENTS.md` → *Pass A*. Work `missing[].suggestions` **and** cross-reference `bedrockOnly[]`,
+since a rename appears on both lists. Verify each candidate against `packs.json` and
+`<section>/files/` before writing to `aliases.json`, then rerun `diff.mjs` and confirm both counts
+moved.
+
+Report aliases added, candidates rejected and why, and any `packs.json` defects found.
+
+## 4. Feasibility pass
+
+```sh
+node tools/vt-diff/feasibility.mjs
+```
+
+Follow `AGENTS.md` → *Pass B*. Dispatch one agent per file in `out/feasibility-batches/`, running
+them in parallel. Each agent researches only its own batch and returns a JSON fragment. Sources, in
+order of preference: the `microsoft-learn` MCP server, `Mojang/bedrock-samples`,
+`wiki.bedrock.dev/llms.txt` and its raw Markdown pages.
+
+Merge the fragments into `feasibility.json`, then:
+
+```sh
+node tools/vt-diff/feasibility.mjs --validate
+```
+
+Fix every reported problem. `unknown` is a valid verdict; a guess is not.
+
+If the developer asked only for an issue refresh and not a research pass, skip this step and say
+that the coverage number in the plan will be low.
+
+## 5. Build and review the issue plan
+
+```sh
+node tools/vt-diff/issues.mjs
+```
+
+Follow `AGENTS.md` → *Pass C*. Verify `reopenCandidates` against `<section>/files/` before reporting
+them — do not assume.
+
+## 6. Stop
+
+Present the plan and ask the developer to approve. Only they run `--apply`.
+
+If `gh auth status` fails, tell them to run `gh auth login` — dry runs work without it, `--apply`
+does not.
+
+## Reporting rules
+
+Lead with the counts table from `report.md` and the feasibility coverage line from `issue-plan.md`.
+Then, in order: newly missing packs since the last run, `reopenCandidates` you verified, `blocked`
+verdicts recorded this run, and category drift. Leave `bedrockOnly` out unless asked — those are our
+originals, not a gap.

.mcp.json

@@ -0,0 +1,8 @@
+{
+	"mcpServers": {
+		"microsoft-learn": {
+			"type": "http",
+			"url": "https://learn.microsoft.com/api/mcp"
+		}
+	}
+}

package.json

@@ -11,7 +11,11 @@
 		"regolith-install": "yarn workspaces foreach -i -A --topological --jobs=1 run regolith-install",
 		"build": "yarn workspaces foreach -i -A run build",
 		"lint": "yarn workspaces foreach -i -A run lint",
-		"test": "yarn workspaces foreach -i -A --continue run test"
+		"test": "yarn workspaces foreach -i -A --continue run test",
+		"vt:fetch": "node tools/vt-diff/fetch.mjs",
+		"vt:diff": "node tools/vt-diff/fetch.mjs && node tools/vt-diff/diff.mjs",
+		"vt:feasibility": "node tools/vt-diff/feasibility.mjs",
+		"vt:issues": "node tools/vt-diff/issues.mjs"
 	},
 	"engines": {
 		"node": ">=25.0.0"