Merge pull request #66 from ThomasByr/feat/refresh-model-catalog-skill

chore(QoL): refresh model catalog skill # Conflicts: # CHANGELOG.md
2026-09-01 23:27:40 +02:00
parent 090d5e5310 8b5e4d78ba
commit 09faff6e6e
6 changed files with 184 additions and 5 deletions
@@ -0,0 +1,76 @@
+---
+name: refresh-model-catalog
+description: Use when adding or removing Command Code models, refreshing the model catalog snapshot (image, reasoning, effort, output-limit metadata), updating display pricing, or refreshing test fixtures in pi-commandcode-provider.
+---
+
+# Refresh Model Catalog
+
+Use this skill whenever the Command Code model catalog changes: new or retired models, changed reasoning efforts, output limits, or pricing. All commands run from the repository root and work on Windows and Linux.
+
+## Core rules
+
+- Do not commit, tag, push, or publish unless the user explicitly asks in the current conversation.
+- Pricing is manually reviewed: temporary promotions and long-context tiers require explicit review of the official pricing page. Never copy prices blindly from the API.
+- Keep the change focused: one refresh per PR, no unrelated refactors.
+- Follow [CONTRIBUTING.md](../../../CONTRIBUTING.md) for commit message rules.
+
+## Workflow
+
+### 1. Detect drift
+
+```sh
+npm run check:commandcode-catalog
+```
+
+This compares the repository snapshot against the latest published `command-code` npm package and reports added/removed models, changed efforts, and version drift. Use the report to scope the work.
+
+### 2. Sync static model metadata
+
+```sh
+npm run sync:commandcode-catalog
+```
+
+Regenerates `src/commandcode-catalog.ts` and bumps the documented CLI version in `README.md`. Review the diff; the catalog also lists reasoning models without selectable efforts.
+
+### 3. Update display pricing (manual review)
+
+Fetch <https://commandcode.ai/docs/resources/pricing-limits> and compare against `src/pricing.ts`:
+
+- Add entries for new models and remove entries for retired models. Missing models silently display zero cost, so `MODEL_COSTS` must cover the full catalog.
+- The pricing page's "Cache Read"/"Cache Write" columns map to `cacheRead`/`cacheWrite`; a "—" column means `0`.
+- Update `PRICING_LAST_VERIFIED` to today's date.
+- Add or update `TEMPORARY_PRICING` entries for promotions with an end date, so tests fail when they expire.
+
+### 4. Refresh the test fixtures
+
+```sh
+node .agents/skills/refresh-model-catalog/scripts/refresh-model-ids.mjs
+npx tsx .agents/skills/refresh-model-catalog/scripts/sync-pricing-fixture.ts
+```
+
+The first script snapshots the live model-id list into `tests/fixtures/commandcode-model-ids.json`; the second regenerates `tests/fixtures/commandcode-pricing.json` from `MODEL_COSTS`. The pricing test fails until `MODEL_COSTS` matches the catalog snapshot exactly.
+
+### 5. Update test expectations
+
+Adjust the model-specific assertions that the refresh invalidated, typically in:
+
+- `tests/test-pricing.ts`: fixture date assertions, the `freeModels` set, and per-model rate assertions.
+- `tests/test-models.ts`: image/reasoning/effort/output-limit assertions and catalog entry counts.
+
+Do not weaken assertions to make them pass; update them to the verified upstream values.
+
+### 6. Validate
+
+```sh
+npm run test:models
+npm run test:pricing
+npm run typecheck
+npm run format:check
+git diff --check
+```
+
+Run the full `npm test` before reporting the work as done when the environment allows it.
+
+### 7. Document
+
+Add entries to the `Unreleased` section of `CHANGELOG.md` covering new/retired models, effort changes, and pricing refreshes.
@@ -0,0 +1,40 @@
+#!/usr/bin/env node
+// Refreshes tests/fixtures/commandcode-model-ids.json from the live Command Code
+// models API. Run from the repository root:
+//   node .agents/skills/refresh-model-catalog/scripts/refresh-model-ids.mjs
+import { writeFile } from "node:fs/promises"
+
+import { format, resolveConfig } from "prettier"
+
+const MODELS_URL = "https://api.commandcode.ai/provider/v1/models"
+const FIXTURE_PATH = new URL(
+  "../../../../tests/fixtures/commandcode-model-ids.json",
+  import.meta.url,
+)
+
+const response = await fetch(MODELS_URL)
+if (!response.ok) {
+  throw new Error(`Failed to fetch Command Code models: ${response.status} ${response.statusText}`)
+}
+
+const body = await response.json()
+if (body?.object !== "list" || !Array.isArray(body.data)) {
+  throw new Error("Expected a Command Code models list response")
+}
+
+const modelIds = body.data.map((model) => {
+  if (typeof model?.id !== "string" || model.id.length === 0) {
+    throw new Error("Expected each model entry to have a non-empty id")
+  }
+  return model.id
+})
+if (modelIds.length === 0) throw new Error("Command Code returned an empty model catalog")
+
+const fixture = { fetchedAt: new Date().toISOString(), source: MODELS_URL, modelIds }
+const options = await resolveConfig(new URL("../../../../.prettierrc.json", import.meta.url))
+const contents = await format(JSON.stringify(fixture), {
+  ...options,
+  filepath: "commandcode-model-ids.json",
+})
+await writeFile(FIXTURE_PATH, contents, "utf-8")
+console.log(`Wrote ${modelIds.length} model ids to tests/fixtures/commandcode-model-ids.json`)
@@ -0,0 +1,43 @@
+// Regenerates tests/fixtures/commandcode-pricing.json from src/pricing.ts so the
+// snapshot always matches MODEL_COSTS. Run from the repository root:
+//   npx tsx .agents/skills/refresh-model-catalog/scripts/sync-pricing-fixture.ts
+import { writeFile } from "node:fs/promises"
+
+import { format, resolveConfig } from "prettier"
+
+import { MODEL_COSTS, PRICING_LAST_VERIFIED, PRICING_SOURCE_URL } from "../../../../src/pricing.ts"
+
+const FIXTURE_PATH = new URL("../../../../tests/fixtures/commandcode-pricing.json", import.meta.url)
+
+const costs: Record<string, [number, number, number, number]> = {}
+const tiers: Record<string, [number, number, number, number, number][]> = {}
+for (const [modelId, cost] of Object.entries(MODEL_COSTS)) {
+  costs[modelId] = [cost.input, cost.output, cost.cacheRead, cost.cacheWrite]
+  if (cost.tiers) {
+    tiers[modelId] = cost.tiers.map((tier) => [
+      tier.inputTokensAbove,
+      tier.input,
+      tier.output,
+      tier.cacheRead,
+      tier.cacheWrite,
+    ])
+  }
+}
+
+const fixture = {
+  verifiedAt: PRICING_LAST_VERIFIED,
+  source: PRICING_SOURCE_URL,
+  tierPolicy:
+    "Use request-wide input tiers; the highest threshold exceeded by input plus cache tokens applies to the full request.",
+  tiers,
+  costs,
+}
+const options = await resolveConfig(new URL("../../../../.prettierrc.json", import.meta.url))
+const contents = await format(JSON.stringify(fixture), {
+  ...options,
+  filepath: "commandcode-pricing.json",
+})
+await writeFile(FIXTURE_PATH, contents, "utf-8")
+console.log(
+  `Wrote ${Object.keys(costs).length} model prices to tests/fixtures/commandcode-pricing.json`,
+)
@@ -18,6 +18,26 @@ const MODELS_REFERENCE_PATH = "dist/bundled/command-code-knowledge/reference/mod
 const CLI_BUNDLE_PATH = "dist/cli.mjs"
 const TEXT_ONLY_MARKER = ',__name(isKnownTextOnlyModel,"isKnownTextOnlyModel")'
 const VALID_EFFORTS = new Set(["minimal", "low", "medium", "high", "xhigh", "max"])
+
+function quoteWindowsArgument(argument: string): string {
+  if (argument.length === 0) return '""'
+  if (!/[\s"]/.test(argument)) return argument
+  return `"${argument.replaceAll('"', '\\"')}"`
+}
+
+/**
+ * Run npm on Windows and POSIX. npm is a `.cmd` shim on Windows, which
+ * execFile cannot spawn directly, so route it through the shell there with
+ * shell-safe quoting.
+ */
+function execNpmFileAsync(
+  args: readonly string[],
+  options: { cwd: string; encoding: "utf-8"; maxBuffer?: number },
+): Promise<{ stdout: string }> {
+  if (process.platform !== "win32") return execFileAsync("npm", args, options)
+  const command = ["npm", ...args.map(quoteWindowsArgument)].join(" ")
+  return execFileAsync(command, { ...options, shell: true })
+}
 const CATALOG_SOURCE_PATH = new URL("../../src/commandcode-catalog.ts", import.meta.url)
 const README_PATH = new URL("../../README.md", import.meta.url)

@@ -417,8 +437,7 @@ async function resolvePackageSpec(
 ): Promise<string> {
  if (packageSpec !== "command-code@latest") return packageSpec

-  const { stdout } = await execFileAsync(
-    "npm",
+  const { stdout } = await execNpmFileAsync(
    ["view", packageSpec, "version", "--json", "--prefer-online", "--cache", npmCacheDirectory],
    {
      cwd: directory,
@@ -437,8 +456,7 @@ async function inspectPackedPackage(packageSpec: string): Promise<{

  try {
    const resolvedPackageSpec = await resolvePackageSpec(packageSpec, directory, npmCacheDirectory)
-    const { stdout } = await execFileAsync(
-      "npm",
+    const { stdout } = await execNpmFileAsync(
      ["pack", resolvedPackageSpec, "--json", "--prefer-online", "--cache", npmCacheDirectory],
      {
        cwd: directory,
@@ -6,6 +6,8 @@
 - Add `Qwen/Qwen3.8-Flash` and `z-ai/glm-5.3-flash` with their verified reasoning efforts (`low, medium, xhigh` and `low, high, max`) and display pricing.
 - Refresh static model capabilities from `command-code@1.36.0`, adding the free `minimax/minimax-m3-free` model and the `z-ai/glm-5.3-flash` output limit while dropping the retired `stealth/ox-alpha`.
 - Refresh display pricing for the current 62-model catalog, adding the free `minimax/minimax-m3-free` and `minimax/minimax-m2.7-free` promotional variants (free through September 5, 2026) and `tencent/hy4-preview`, and removing the retired `stealth/ox-alpha`.
+- Fix `npm run sync:commandcode-catalog` and `npm run check:commandcode-catalog` on Windows by spawning npm through the shell.
+- Add a `refresh-model-catalog` agent skill with cross-platform helper scripts that snapshot the live model catalog and regenerate the pricing fixture from `MODEL_COSTS`.

 ## 0.6.0 - 2026-08-25

@@ -9,5 +9,5 @@
    "strict": true,
    "types": ["node"]
  },
-  "include": [".github/scripts/**/*.ts", "src/**/*.ts", "tests/**/*.ts"]
+  "include": [".github/scripts/**/*.ts", ".agents/skills/**/*.ts", "src/**/*.ts", "tests/**/*.ts"]
 }