feat: recalculate statusline cost from model prices

.claude-plugin/marketplace.json

@@ -15,6 +15,7 @@
       "strict": false,
       "skills": [
         "./skills/webup-buddy-reroll",
+        "./skills/webup-model-price",
         "./skills/webup-statusline"
       ]
     }

.github/workflows/ci.yml

@@ -6,7 +6,7 @@ on:
   pull_request:
 
 jobs:
-  verify-statusline:
+  verify-skills:
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
@@ -16,6 +16,21 @@ jobs:
 
       - uses: oven-sh/setup-bun@v2
 
+      - name: Install jq on Windows
+        if: runner.os == 'Windows'
+        run: |
+          choco install jq -y
+          "C:\ProgramData\chocolatey\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
+        shell: powershell
+
+      - name: Verify model price skill
+        run: python3 scripts/verify_model_price.py
+        shell: bash
+
+      - name: Verify statusline replay fixture
+        run: python3 scripts/verify_statusline_replay.py
+        shell: bash
+
       - name: Verify statusline generation
         run: python3 scripts/verify_context.py
         shell: bash

README.ja.md

@@ -21,6 +21,23 @@ npx skills add webup/skills-cc -s webup-statusline -g
 
 ## 🎮 スキル一覧
 
+### 💵 webup-model-price
+
+[`models.dev`](https://models.dev) からモデル価格を取得して正規化します。provider 指定、複数プラットフォーム比較、catalog キャッシュ、token 数からの費用見積もりに対応します。
+
+```bash
+# Anthropic Claude の価格
+/webup-model-price claude-opus-4-7
+
+# 完全一致する全プロバイダーの価格を比較
+npx -y bun skills/webup-model-price/scripts/price.mjs --model claude-opus-4-7 --all
+
+# token 数から費用を見積もる
+npx -y bun skills/webup-model-price/scripts/price.mjs --provider anthropic --model claude-opus-4-7 --input-tokens 100000 --output-tokens 10000
+```
+
+ステータスラインのコスト再計算でも使えるほか、単体で価格確認や provider 比較にも使えます。
+
 ### 📊 webup-statusline
 
 カスタム Claude Code ステータスラインを生成・インストール。**カラム** と **テーマ** を選ぶだけ。`context` と `effort` の 2 カラムは **レベルに応じて色が変わる** ので、一目で状況が分かります。
@@ -74,7 +91,7 @@ npx skills add webup/skills-cc -s webup-statusline -g
 |--------|----------|----------|
 | `model` | アクティブモデル名 | 常時 |
 | `context` | コンテキスト進捗バー + % — **残量に応じて色が変化** | 常時 |
-| `cost` | セッション API 支出 `$X.XX`（ゴールド、例：`$0.42`） | `cost.total_cost_usd` が四捨五入して ≥ $0.01 の時 |
+| `cost` | 現在モデルの models.dev 単価で再計算したセッション支出 `$X.XX`（ゴールド、例：`$0.42`）。再計算できない場合は Claude Code の `cost.total_cost_usd` にフォールバック | 表示値が四捨五入して ≥ $0.01 の時 |
 | `effort` | 推論努力レベル — **強度で色分け**（`low`/`medium`/`high`/`xhigh`/`max` 対応） | `~/.claude/settings.json` で `effortLevel` が設定されている時 |
 | `style` | 出力スタイル名（例：`Explanatory`、`Learning`）— Claude のブランドカラーに合わせたパープル | `output_style.name` が `default` 以外の時 |
 | `dir` | リポジトリディレクトリ名（ワークツリー内では元リポジトリ名） | 常時 |
@@ -107,7 +124,7 @@ npx skills add webup/skills-cc -s webup-statusline -g
 
 **effort アイコンの上書き**は `--effort-icon` で。プリセット：`arrow`（`↯`、デフォルト）、`bolt`（`ϟ`）、`flash`（`⚡`）、`reason`（`∴`）、`dot`（`◉`）、`none`（非表示）。任意の文字も受け付けます。
 
-> ⚠️ **注意：** 生成されたスクリプトは JSON 解析に `jq` が必要です。Windows では WinGet や scoop でインストールされた jq のパスを自動検出します。それでも jq が見つからない場合は、手動でディレクトリを PATH に追加してください。スキルは `~/.claude/scripts/statusline.sh` を自動生成し `~/.claude/settings.json` を更新します。Claude Code を再起動すると反映されます。
+> ⚠️ **注意：** 生成されたスクリプトは JSON 解析に `jq` が必要です。Windows では WinGet、Chocolatey、scoop でインストールされた jq のパスを自動検出します。それでも jq が見つからない場合は、手動でディレクトリを PATH に追加してください。スキルは `~/.claude/scripts/statusline.sh` を自動生成し `~/.claude/settings.json` を更新します。Claude Code を再起動すると反映されます。`cost` カラムは `prompt_id` ごとの再計算結果をローカル状態ディレクトリに記録し、ステータスラインの再描画で二重計上しないようにします。
 
 ### 🎰 webup-buddy-reroll
 

README.md

@@ -21,6 +21,23 @@ npx skills add webup/skills-cc -s webup-statusline -g
 
 ## 🎮 Skills
 
+### 💵 webup-model-price
+
+Look up normalized model prices from [`models.dev`](https://models.dev), with provider-aware search, cached catalog data, and token-cost estimation.
+
+```bash
+# Anthropic-hosted Claude price
+/webup-model-price claude-opus-4-7
+
+# Compare all providers/platforms with an exact match
+npx -y bun skills/webup-model-price/scripts/price.mjs --model claude-opus-4-7 --all
+
+# Estimate usage cost from token counts
+npx -y bun skills/webup-model-price/scripts/price.mjs --provider anthropic --model claude-opus-4-7 --input-tokens 100000 --output-tokens 10000
+```
+
+This skill powers the statusline cost recalculation path while staying useful on its own for model price checks and provider comparisons.
+
 ### 📊 webup-statusline
 
 Generate and install a custom Claude Code status line. Pick your **columns**, pick a **theme** — that's it. Two columns (`context` and `effort`) also **change color based on level**, so the bar tells you at a glance where you stand.
@@ -74,7 +91,7 @@ Healthy session — 88% remaining, `effortLevel: medium`, default output style (
 |--------|---------------|--------------|
 | `model` | Active model name | Always |
 | `context` | Context window progress bar + percentage — **color scales with remaining capacity** | Always |
-| `cost` | Session API spend as `$X.XX` in gold (e.g. `$0.42`) | When `cost.total_cost_usd` rounds to ≥ $0.01 |
+| `cost` | Recalculated session spend as `$X.XX` in gold (e.g. `$0.42`) using the current model's models.dev price, with Claude Code's `cost.total_cost_usd` as fallback | When the displayed value rounds to ≥ $0.01 |
 | `effort` | Reasoning effort level — **colored by intensity** (supports `low`/`medium`/`high`/`xhigh`/`max`) | When `effortLevel` is set in `~/.claude/settings.json` |
 | `style` | Output style name (e.g. `Explanatory`, `Learning`) — shown in purple to match Claude's brand hue | When `output_style.name` is anything other than `default` |
 | `dir` | Repo directory basename (original repo when inside a worktree) | Always |
@@ -107,7 +124,7 @@ The `context` column intentionally has no prefix icon in any theme — the color
 
 **Override the effort icon** via `--effort-icon`. Presets: `arrow` (`↯`, default), `bolt` (`ϟ`), `flash` (`⚡`), `reason` (`∴`), `dot` (`◉`), `none` (hide). A raw character is also accepted.
 
-> ⚠️ **Note:** The generated script requires `jq` for JSON parsing. On Windows, it auto-detects jq installed via WinGet or scoop; if jq is still not found, add its directory to your PATH manually. The skill writes to `~/.claude/scripts/statusline.sh` and updates `~/.claude/settings.json` — restart Claude Code to see it.
+> ⚠️ **Note:** The generated script requires `jq` for JSON parsing. On Windows, it auto-detects jq installed via WinGet, Chocolatey, or scoop; if jq is still not found, add its directory to your PATH manually. The skill writes to `~/.claude/scripts/statusline.sh` and updates `~/.claude/settings.json` — restart Claude Code to see it.
 
 ### 🎰 webup-buddy-reroll
 

README.zh.md

@@ -21,6 +21,23 @@ npx skills add webup/skills-cc -s webup-statusline -g
 
 ## 🎮 技能列表
 
+### 💵 webup-model-price
+
+从 [`models.dev`](https://models.dev) 查询并标准化模型价格，支持指定 provider、跨平台比较、缓存 catalog，以及根据 token 数估算费用。
+
+```bash
+# 查询 Anthropic Claude 价格
+/webup-model-price claude-opus-4-7
+
+# 查看所有精确匹配的平台价格
+npx -y bun skills/webup-model-price/scripts/price.mjs --model claude-opus-4-7 --all
+
+# 根据 token 数估算费用
+npx -y bun skills/webup-model-price/scripts/price.mjs --provider anthropic --model claude-opus-4-7 --input-tokens 100000 --output-tokens 10000
+```
+
+这个技能会被状态栏的费用重算逻辑复用，也可以单独用于查价和比较平台价格。
+
 ### 📊 webup-statusline
 
 生成并安装自定义 Claude Code 状态栏。选择 **字段** 和 **主题**，仅此而已。其中两个字段（`context` 和 `effort`）会 **随级别动态换色**，一眼看清当前状态。
@@ -74,7 +91,7 @@ npx skills add webup/skills-cc -s webup-statusline -g
 |------|----------|----------|
 | `model` | 活跃模型名 | 始终显示 |
 | `context` | 上下文进度条 + 百分比 —— **颜色随剩余容量变化** | 始终显示 |
-| `cost` | 会话累计消费 `$X.XX`（金色，例如 `$0.42`） | 当 `cost.total_cost_usd` 四舍五入后 ≥ $0.01 |
+| `cost` | 按当前模型的 models.dev 单价重算会话消费 `$X.XX`（金色，例如 `$0.42`），无法重算时回退到 Claude Code 的 `cost.total_cost_usd` | 当显示值四舍五入后 ≥ $0.01 |
 | `effort` | 推理努力度 —— **按强度着色**（支持 `low`/`medium`/`high`/`xhigh`/`max`） | 当 `~/.claude/settings.json` 中设置了 `effortLevel` |
 | `style` | 输出样式名（例如 `Explanatory`、`Learning`）—— 紫色呼应 Claude 品牌色 | 当 `output_style.name` 不是 `default` 时 |
 | `dir` | 仓库目录名（在工作树中显示原仓库名） | 始终显示 |
@@ -107,7 +124,7 @@ npx skills add webup/skills-cc -s webup-statusline -g
 
 **替换 effort 图标** 用 `--effort-icon`。预设值：`arrow`（`↯`，默认）、`bolt`（`ϟ`）、`flash`（`⚡`）、`reason`（`∴`）、`dot`（`◉`）、`none`（隐藏）。也可直接传入任意字符。
 
-> ⚠️ **注意：** 生成的脚本需要 `jq` 解析 JSON。在 Windows 下，脚本会自动检测 WinGet 和 scoop 安装的 jq 路径；若仍无法找到 jq，请手动将其目录添加到 PATH。本技能会自动写入 `~/.claude/scripts/statusline.sh` 并更新 `~/.claude/settings.json` —— 重启 Claude Code 即可生效。
+> ⚠️ **注意：** 生成的脚本需要 `jq` 解析 JSON。在 Windows 下，脚本会自动检测 WinGet、Chocolatey 和 scoop 安装的 jq 路径；若仍无法找到 jq，请手动将其目录添加到 PATH。本技能会自动写入 `~/.claude/scripts/statusline.sh` 并更新 `~/.claude/settings.json` —— 重启 Claude Code 即可生效。`cost` 字段会把每个 `prompt_id` 的重算费用记录到本地状态目录，避免状态栏重复刷新时重复计费。
 
 ### 🎰 webup-buddy-reroll
 

scripts/verify_context.py

@@ -5,6 +5,7 @@
 import sys
 import tempfile
 import os
+import json
 
 THEMES = ["gruvbox", "dracula", "robbyrussell", "minimal"]
 ELEMENT_SETS = [
@@ -13,6 +14,7 @@
     "model,context,cost,effort,style,git,dir,worktree,vim",
 ]
 GENERATOR = "skills/webup-statusline/scripts/generate.mjs"
+PRICE_FIXTURE = os.path.abspath("tests/fixtures/models-dev-catalog.json")
 
 IS_WINDOWS = sys.platform == "win32"
 
@@ -45,6 +47,106 @@ def bash_syntax_ok(script):
         os.unlink(path)
 
 
+def run_statusline_script(script, payload, env):
+    fd, path = tempfile.mkstemp(suffix=".sh")
+    try:
+        with os.fdopen(fd, "w", encoding="utf-8", newline="\n") as f:
+            f.write(script)
+        return subprocess.run(
+            ["bash", path],
+            input=json.dumps(payload),
+            capture_output=True,
+            encoding="utf-8",
+            errors="replace",
+            env={**os.environ, **env},
+        )
+    finally:
+        os.unlink(path)
+
+
+def statusline_payload(model_id, display_name, prompt_id):
+    return {
+        "session_id": "verify-model-switch",
+        "prompt_id": prompt_id,
+        "model": {
+            "id": model_id,
+            "display_name": display_name,
+        },
+        "workspace": {
+            "current_dir": os.getcwd(),
+        },
+        "cost": {
+            "total_cost_usd": 0.01,
+        },
+        "context_window": {
+            "remaining_percentage": 90,
+            "total_input_tokens": 100000,
+            "total_output_tokens": 10000,
+            "context_window_size": 200000,
+            "current_usage": {
+                "input_tokens": 100000,
+                "output_tokens": 10000,
+                "cache_creation_input_tokens": 0,
+                "cache_read_input_tokens": 0,
+            },
+        },
+    }
+
+
+def model_switch_cost_ok(script):
+    """Verify cost is recomputed from per-model prices and repeated refreshes do not double-count."""
+    if IS_WINDOWS:
+        return True, "(skipped on Windows — Linux/macOS CI covers execution)"
+
+    with tempfile.TemporaryDirectory() as state_dir:
+        env = {
+            "WEBUP_MODEL_PRICE_CATALOG": PRICE_FIXTURE,
+            "WEBUP_STATUSLINE_STATE_DIR": state_dir,
+        }
+
+        missing = run_statusline_script(
+            script,
+            statusline_payload("unknown-model", "Unknown Model", "prompt-missing"),
+            env,
+        )
+        if missing.returncode != 0:
+            return False, f"missing-price fallback render failed: {missing.stderr.strip()}"
+        if "$0.01" not in missing.stdout:
+            return False, f"missing-price fallback expected Claude Code $0.01, got: {missing.stdout!r}"
+
+        first = run_statusline_script(
+            script,
+            statusline_payload("claude-sonnet-4-6", "Claude Sonnet 4.6", "prompt-1"),
+            env,
+        )
+        if first.returncode != 0:
+            return False, f"first render failed: {first.stderr.strip()}"
+        if "$0.45" not in first.stdout:
+            return False, f"first render expected $0.45, got: {first.stdout!r}"
+
+        second = run_statusline_script(
+            script,
+            statusline_payload("claude-opus-4-7", "Claude Opus 4.7", "prompt-2"),
+            env,
+        )
+        if second.returncode != 0:
+            return False, f"second render failed: {second.stderr.strip()}"
+        if "$1.20" not in second.stdout:
+            return False, f"second render expected $1.20, got: {second.stdout!r}"
+
+        repeat = run_statusline_script(
+            script,
+            statusline_payload("claude-opus-4-7", "Claude Opus 4.7", "prompt-2"),
+            env,
+        )
+        if repeat.returncode != 0:
+            return False, f"repeat render failed: {repeat.stderr.strip()}"
+        if "$1.20" not in repeat.stdout:
+            return False, f"repeat render expected $1.20 without double-counting, got: {repeat.stdout!r}"
+
+    return True, ""
+
+
 def main():
     errors = []
     for theme in THEMES:
@@ -92,6 +194,11 @@ def main():
                 if el == "vim" and "vim.mode" not in script:
                     errors.append(f"FAIL missing vim field in {label}")
 
+            if theme == "minimal" and elements == "model,context,cost,effort,style,git,dir,worktree,vim":
+                ok, detail = model_switch_cost_ok(script)
+                if not ok:
+                    errors.append(f"FAIL model switch cost ledger {label}: {detail}")
+
     if errors:
         for e in errors:
             print(e, file=sys.stderr)

scripts/verify_model_price.py

@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+"""Verify models.dev price lookup behavior."""
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+PRICE = ROOT / "skills/webup-model-price/scripts/price.mjs"
+FIXTURE = ROOT / "tests/fixtures/models-dev-catalog.json"
+
+
+def run_price(*args):
+    return subprocess.run(
+        ["bun", str(PRICE), "--catalog", str(FIXTURE), *args],
+        capture_output=True,
+        encoding="utf-8",
+        errors="replace",
+    )
+
+
+def assert_ok(result, label):
+    if result.returncode != 0:
+      raise AssertionError(f"{label} failed:\nSTDOUT={result.stdout}\nSTDERR={result.stderr}")
+
+
+def main():
+    errors = []
+
+    try:
+        r = run_price("--provider", "anthropic", "--model", "claude-opus-4-7")
+        assert_ok(r, "anthropic exact lookup")
+        data = json.loads(r.stdout)
+        assert data["provider"] == "anthropic"
+        assert data["model"] == "claude-opus-4-7"
+        assert data["costPerMillion"] == {
+            "input": 5,
+            "output": 25,
+            "cacheRead": 0.5,
+            "cacheWrite": 6.25,
+        }
+    except Exception as exc:
+        errors.append(f"FAIL exact provider lookup: {exc}")
+
+    try:
+        r = run_price("--model", "claude-opus-4-7", "--all")
+        assert_ok(r, "all provider lookup")
+        data = json.loads(r.stdout)
+        providers = [item["provider"] for item in data["matches"]]
+        assert providers == ["anthropic", "openrouter", "mirror"], providers
+    except Exception as exc:
+        errors.append(f"FAIL all provider lookup: {exc}")
+
+    try:
+        r = run_price("--model", "anthropic/claude-opus-4.7", "--all")
+        assert_ok(r, "provider-prefixed lookup")
+        data = json.loads(r.stdout)
+        assert [item["provider"] for item in data["matches"]] == ["openrouter"]
+    except Exception as exc:
+        errors.append(f"FAIL provider-prefixed lookup: {exc}")
+
+    try:
+        r = run_price("--model", "claude-opus-4-7", "--no-default-provider")
+        if r.returncode == 0:
+            raise AssertionError("ambiguous lookup unexpectedly succeeded")
+        if "ambiguous" not in r.stderr.lower() or "anthropic" not in r.stderr or "openrouter" not in r.stderr:
+            raise AssertionError(f"unexpected ambiguity message: {r.stderr}")
+    except Exception as exc:
+        errors.append(f"FAIL ambiguous lookup: {exc}")
+
+    try:
+        r = run_price(
+            "--provider", "anthropic",
+            "--model", "claude-opus-4-7",
+            "--input-tokens", "100000",
+            "--output-tokens", "10000",
+            "--cache-read-tokens", "20000",
+            "--cache-write-tokens", "5000",
+        )
+        assert_ok(r, "cost estimate")
+        data = json.loads(r.stdout)
+        assert data["estimatedCostUsd"] == 0.79125, data["estimatedCostUsd"]
+    except Exception as exc:
+        errors.append(f"FAIL cost estimate: {exc}")
+
+    if errors:
+        for error in errors:
+            print(error, file=sys.stderr)
+        print(f"\n{len(errors)} check(s) failed", file=sys.stderr)
+        sys.exit(1)
+
+    print("Model price lookup checks passed")
+
+
+if __name__ == "__main__":
+    main()