coderouter-cli 1.8.3__py3-none-any.whl → 1.8.5__py3-none-any.whl

coderouter/doctor.py

@@ -836,6 +836,22 @@ async def _probe_num_ctx(
         )
 
     # Canary missing → truncation occurred.
+    #
+    # v1.8.5: with the v1.8.3 thinking-aware response budget already
+    # applied (max_tokens=1024 for reasoning models — see
+    # ``_NUM_CTX_PROBE_MAX_TOKENS_THINKING``), a missing canary cannot
+    # be blamed on an under-sized reply. The fault is genuinely on the
+    # prompt side: the upstream truncated the input before the model
+    # could see the canary token at the head. This sharpens the
+    # remediation — we are confident bumping ``num_ctx`` is the right
+    # fix, not bumping the response budget.
+    thinking = _is_reasoning_model(provider, resolved)
+    budget_note = (
+        f" Probe sent max_tokens={max_tokens} (thinking-aware), so the "
+        "miss is prompt-side truncation rather than reply truncation."
+        if thinking
+        else ""
+    )
     if declared is None:
         return ProbeResult(
             name="num_ctx",
@@ -845,7 +861,7 @@ async def _probe_num_ctx(
                 "upstream truncated the prompt. No `extra_body.options.num_ctx` "
                 "is declared, so Ollama is running at its 2048-token default, "
                 "which cannot hold Claude Code's system + tool prompts "
-                "(plan.md §9.4 symptom #1)."
+                f"(plan.md §9.4 symptom #1).{budget_note}"
             ),
             target_file="providers.yaml",
             suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
@@ -857,7 +873,8 @@ async def _probe_num_ctx(
             detail=(
                 f"canary missing — declared num_ctx={declared} is below "
                 f"the {_NUM_CTX_ADEQUATE_THRESHOLD}-token threshold needed "
-                "for Claude Code prompts. Bump it (plan.md §9.4 symptom #1)."
+                f"for Claude Code prompts. Bump it (plan.md §9.4 symptom "
+                f"#1).{budget_note}"
             ),
             target_file="providers.yaml",
             suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
@@ -875,7 +892,7 @@ async def _probe_num_ctx(
             "declared value, or the upstream is silently capping it — "
             "verify with the model card / server logs. The suggested "
             "patch still emits 32768 as a starting point; dial down if "
-            "the host is memory-constrained."
+            f"the host is memory-constrained.{budget_note}"
         ),
         target_file="providers.yaml",
         suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
@@ -1041,6 +1058,24 @@ async def _probe_streaming(
         # mid-word". Since we're already Ollama-shape-gated, the
         # remediation is always the ``extra_body.options.num_predict``
         # bump.
+        #
+        # v1.8.5: with v1.8.3's thinking-aware probe budget already
+        # applied (max_tokens=1024 for reasoning models), a length cap
+        # here cannot be blamed on the probe budget — the upstream is
+        # the one capping. Surface the budget used so the operator can
+        # rule it out at a glance.
+        thinking = _is_reasoning_model(provider, resolved)
+        budget_note = (
+            f"Probe sent max_tokens={max_tokens} (thinking-aware), so "
+            "the cap is server-side `options.num_predict` rather than "
+            "the probe budget."
+            if thinking
+            else (
+                f"Probe sent max_tokens={max_tokens}; the cap is "
+                "server-side `options.num_predict` rather than the "
+                "probe budget."
+            )
+        )
         return ProbeResult(
             name="streaming",
             verdict=ProbeVerdict.NEEDS_TUNING,
@@ -1048,9 +1083,9 @@ async def _probe_streaming(
                 f"stream closed with `finish_reason='length'` after only "
                 f"{len(content)} chars (expected ≥ "
                 f"{_STREAMING_PROBE_MIN_EXPECTED_CHARS}). Upstream is "
-                "capping output — most likely `options.num_predict`. "
-                "Bump it via `extra_body` (plan.md §9.4 symptom #1 "
-                "streaming variant)."
+                f"capping output — most likely `options.num_predict`. "
+                f"{budget_note} Bump it via `extra_body` (plan.md §9.4 "
+                "symptom #1 streaming variant)."
             ),
             target_file="providers.yaml",
             suggested_patch=_patch_providers_yaml_num_predict(
@@ -1239,13 +1274,33 @@ async def _probe_tool_calls(
 
     # Nothing tool-shaped at all.
     if declared:
+        # v1.8.5: with the v1.8.3 thinking-aware budget already applied,
+        # we can speak with confidence here: the model genuinely did not
+        # emit tool_calls (it's not a budget-exhaustion false-positive
+        # like the pre-v1.8.3 64-token cap used to produce). For thinking
+        # models specifically, the 1024-token budget covers
+        # ``reasoning_content`` *and* a tool call — so a missing
+        # ``tool_calls`` here is real. Surface the budget that was used
+        # so operators reading the message understand what was probed.
+        thinking = _is_reasoning_model(provider, resolved)
+        budget_note = (
+            f"Probed with thinking-aware budget ({max_tokens} tokens, "
+            "covers `reasoning_content` plus the call) — this is a true "
+            "tools=false case, not budget exhaustion."
+            if thinking
+            else (
+                f"Probed with default budget ({max_tokens} tokens) — "
+                "the model produced no tool-shaped output at all."
+            )
+        )
         return ProbeResult(
             name="tool_calls",
             verdict=ProbeVerdict.NEEDS_TUNING,
             detail=(
                 "declaration says tools=true but model produced neither "
-                "native `tool_calls` nor repairable tool JSON. Common for "
-                "quantized small models (plan.md §9.4 symptom #2)."
+                "native `tool_calls` nor repairable tool JSON. "
+                f"{budget_note} Common for quantized small models "
+                "(plan.md §9.4 symptom #2)."
             ),
             target_file="providers.yaml",
             suggested_patch=_patch_providers_yaml_capability(provider.name, "tools", False),

{coderouter_cli-1.8.3.dist-info → coderouter_cli-1.8.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coderouter-cli
-Version: 1.8.3
+Version: 1.8.5
 Summary: Local-first, free-first, fallback-built-in LLM router. Claude Code / OpenAI compatible.
 Project-URL: Homepage, https://github.com/zephel01/CodeRouter
 Project-URL: Repository, https://github.com/zephel01/CodeRouter
@@ -60,7 +60,7 @@ Description-Content-Type: text/markdown
 <p align="center">
   <a href="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml"><img src="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
   <a href=""><img src="https://img.shields.io/badge/status-stable-brightgreen" alt="status"></a>
-  <a href=""><img src="https://img.shields.io/badge/version-1.8.3-blue" alt="version"></a>
+  <a href=""><img src="https://img.shields.io/badge/version-1.8.5-blue" alt="version"></a>
   <a href=""><img src="https://img.shields.io/badge/python-3.12%2B-blue" alt="python"></a>
   <a href=""><img src="https://img.shields.io/badge/runtime%20deps-5-brightgreen" alt="deps"></a>
   <a href=""><img src="https://img.shields.io/badge/license-MIT-yellow" alt="license"></a>
@@ -99,11 +99,13 @@ Description-Content-Type: text/markdown
 | **無料で回す** | [無料枠ガイド](./docs/free-tier-guide.md) | NVIDIA NIM 40 req/min × OpenRouter 無料枠の使い分け・live 検証済みモデル表・地雷 5 点 |
 | **要るか判断する** | [要否判定ガイド](./docs/when-do-i-need-coderouter.md) | エージェント × モデルの詳細マトリクスで「そもそも自分に必要か」を決める |
 | **詰まったとき** | [トラブルシューティング](./docs/troubleshooting.md) | `doctor` の使い方、`.env` の export 必須、Ollama サイレント失敗 5 症状、Claude Code 連携の罠 |
+| **llama.cpp 直叩き** | [llama.cpp 直叩きガイド](./docs/llamacpp-direct.md) | Qwen3.6 を Ollama 詰みから救出する経路。`llama.cpp` build → Unsloth GGUF → `llama-server` → CodeRouter 接続を 7 step で（v1.8.3 実機検証済）|
+| **LM Studio 直接** | [LM Studio 直接ガイド](./docs/lmstudio-direct.md) | `qwen35` / `qwen35moe` を救う第 2 経路。LM Studio 0.4.12+ Local Server 経由で OpenAI 互換 + Anthropic 互換 (`/v1/messages`) 両対応、prompt caching 透過（v1.8.4 実機検証済）|
 | **安全に使う** | [セキュリティ方針](./docs/security.md) | 脅威モデル・秘密情報の扱い・脆弱性報告経路 |
-| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴（最新: v1.8.3 — tool_calls probe も thinking 対応 + adapter で `reasoning_content` strip / llama.cpp 直叩き対応） |
+| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴（最新: v1.8.5 — doctor NEEDS_TUNING メッセージを v1.8.3 thinking-aware budget の事実に揃える + `docs/lmstudio-direct.md` 新規） |
 | **設計を追う** | [plan.md](./plan.md) | 設計不変項・マイルストーン・今後のロードマップ |
 
-English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [Security](./docs/security.en.md)
+English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [llama.cpp direct](./docs/llamacpp-direct.en.md) · [LM Studio direct](./docs/lmstudio-direct.en.md) · [Security](./docs/security.en.md)
 
 ## CodeRouter で何が楽になるか
 
@@ -184,7 +186,9 @@ curl -fsSL https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/p
   > ~/.coderouter/providers.yaml
 
 # 2. uvx で起動 (インストール + 起動が 1 行)
-uvx coderouter-cli serve --port 8088
+#    PyPI 配布名 (coderouter-cli) と console script 名 (coderouter) が異なるため、
+#    uv 0.11+ では --from 形式が必須 (旧 uv でも動く canonical 形式)
+uvx --from coderouter-cli coderouter serve --port 8088
 ```
 
 恒久的にインストールしておきたい場合:
@@ -257,7 +261,7 @@ CodeRouter 自体は純 Python 3.12+ で、実質的な OS 対応範囲は `min(
 
 **リリース単位の詳細が欲しい？** v0.x と v1.0-A/B/C の各スライス — 何が入り、何本のテストが増え、なぜ必要だったのか — は [CHANGELOG.md](./CHANGELOG.md) に揃っています。設計の不変項と今後のロードマップは [plan.md](./plan.md)。
 
-**次の予定**（v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18）: v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx coderouter-cli`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
+**次の予定**（v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18）: v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx --from coderouter-cli coderouter`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
 
 ### Claude Code と一緒に使う
 
@@ -432,7 +436,7 @@ suggested patch for ~/.coderouter/providers.yaml:
 - v1.0 ✅ — 14 ケースのリグレッションスイート、Code Mode (スリム版 Claude Code ハーネス); 出力クリーニングは **v1.0-A** で `output_filters` チェーンとして完了
 - v1.5 ✅ — **メトリクスダッシュボード（出荷済み）** — `MetricsCollector` + `GET /metrics.json` + `GET /metrics` (Prometheus) + `GET /dashboard` (HTML 1 ページ) + `coderouter stats` curses TUI + `scripts/demo_traffic.sh` トラフィックジェネレータ + `display_timezone` 設定
 - v1.6 ✅ — `auto_router` (task-aware routing、`default_profile: auto` で画像/コード濃度/その他を自動振り分け) + NVIDIA NIM 無料枠 8 段チェーン + ドキュメント言語スワップ (JA primary) + トラブルシュート独立ドキュメント + `--env-file` / `doctor --check-env`
-- v1.7 ✅ — PyPI 公開 (`uvx coderouter-cli` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
+- v1.7 ✅ — PyPI 公開 (`uvx --from coderouter-cli coderouter` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
 - v1.8 ✅ — **用途別 4 プロファイル + GLM/Gemma 4/Qwen3.6 公式化 + apply 自動化**: `multi` (default) / `coding` / `general` / `reasoning` の 4 プロファイル + 全プロファイルに `append_system_prompt` で Claude 風応答 nudge + `mode_aliases` (default/fast/vision/think/cheap)、Ollama 公式 tag 化された `gemma4:e4b/26b/31b` / `qwen3.6:27b/35b` を active stanza に格上げ、Z.AI を OpenAI-compat で 2 base_url 提供 (Coding Plan / General API)、`coderouter doctor --check-model --apply` で YAML パッチを非破壊書き戻し (`ruamel.yaml` round-trip でコメント・key 順序保持、冪等)、`setup.sh` onboarding ウィザード、`claude_code_suitability` startup チェック (Llama-3.3-70B 系を `claude-code-*` profile で WARN)。残り (v1.9 以降): `coderouter doctor --network` (CI 用)、launcher スクリプト (`.command` / `.sh` / `.bat`)、opt-in 起動時アップデートチェック
 
 ## `kind: openai_compat` と `kind: anthropic` の選び方

{coderouter_cli-1.8.3.dist-info → coderouter_cli-1.8.5.dist-info}/RECORD

@@ -2,7 +2,7 @@ coderouter/__init__.py,sha256=ghdjPrLtnRzY8fyQ4CJZI1UJKADyNTLtA3G7se8H7Ns,696
 coderouter/__main__.py,sha256=-LCgxJnvgUV240HjQKv7ly-mn2NuKHpC4nCpvTHjeSU,130
 coderouter/cli.py,sha256=vI1-dv10t4-xG6Zpt7zi_3U8xGgq54Qa8XIMUYpfOV8,19859
 coderouter/cli_stats.py,sha256=ae20xUr_hjX09Ms3fBZGZsUS52o44JC57EpbWLBOCO0,27750
-coderouter/doctor.py,sha256=F6f9vl99KTGgCja6N9w_QlDIFFCEqY0mxoAhKEI5yTI,69643
+coderouter/doctor.py,sha256=Gs9KIYHvXBwhnNlt5rzx6vceEliV5gJ6cBuf_pMLQ6A,72417
 coderouter/doctor_apply.py,sha256=r_J6xbu5-HivofPNriw4_vjNYs_VRs7GsGTS0oMEX10,24209
 coderouter/env_security.py,sha256=FEBZnXfJ0xE39kmMMn39zk0W_DRRnmcB_REmP9f4xWo,14796
 coderouter/errors.py,sha256=Xmq67lheyw8iv3Ox39jh2c4tvNI5RcUR4QkoxVDN6l4,1130
@@ -37,8 +37,8 @@ coderouter/translation/__init__.py,sha256=PYXN7XVEwpG1uC8RLy6fvnGbzEZhhrEuUapH8I
 coderouter/translation/anthropic.py,sha256=JpvIWNXHUPVqOGvps7o_6ZADhXuJuvpU7RdMqQFtwwM,6421
 coderouter/translation/convert.py,sha256=-qyzFzmmr9hhQV6_Sg75kJnvCZvHe3n7vRdaZtk_JqQ,47269
 coderouter/translation/tool_repair.py,sha256=fyxDb4kWHytO5JWq5y0i4tinJUtWqhMCkyfoCf5BjeM,8314
-coderouter_cli-1.8.3.dist-info/METADATA,sha256=Nm6kOVjXop9D5aISl3H0OcH-gi4fkGT0fmhuyoKTcHU,44221
-coderouter_cli-1.8.3.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
-coderouter_cli-1.8.3.dist-info/entry_points.txt,sha256=-dnLfD1YZ2WjH2zSdNCvlO65wYltM9bsHt9Fhg3yGss,51
-coderouter_cli-1.8.3.dist-info/licenses/LICENSE,sha256=wkEzoR86jFw33jvfOHjULqmkGEfxTFMgMaJnpR8mPRw,1065
-coderouter_cli-1.8.3.dist-info/RECORD,,
+coderouter_cli-1.8.5.dist-info/METADATA,sha256=7WObVqSwWtedbYe0LUFR7ZoXR_r0tQRquER992PFAKM,45119
+coderouter_cli-1.8.5.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+coderouter_cli-1.8.5.dist-info/entry_points.txt,sha256=-dnLfD1YZ2WjH2zSdNCvlO65wYltM9bsHt9Fhg3yGss,51
+coderouter_cli-1.8.5.dist-info/licenses/LICENSE,sha256=wkEzoR86jFw33jvfOHjULqmkGEfxTFMgMaJnpR8mPRw,1065
+coderouter_cli-1.8.5.dist-info/RECORD,,