fix: harden llm reports and model discovery

Author: yjw

CHANGELOG.md

@@ -5,6 +5,19 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.1] - 2026-06-18
+
+### Added
+- Added `young config models` for provider-neutral model discovery across Ark and other OpenAI-compatible services, Kimi/Moonshot, DeepSeek, Qwen, Anthropic, and Ollama endpoints.
+- Added report-time stock-analysis specification checks with safe text-only caching; remote code is never executed.
+
+### Fixed
+- Translated internal Evidence Pack fields into research language before LLM synthesis and sanitized engineering terminology from final reports.
+- Allowed verified fund-flow data to keep the sector/fund-flow module available when board rankings are temporarily absent.
+- Unified `young a` and LLM evidence board routing so both try the lightweight source and then the configured Camofox browser path.
+- Converted Camofox board snapshots into the same structured row format used by normal board rankings.
+- Expanded uv tool installation and PDF dependency guidance in README and command errors.
+
 ## [0.2.0] - 2026-06-18
 
 ### Added

README.md

@@ -17,6 +17,14 @@ Born out of a real workflow: every trading day after close I wanted the same fiv
 
 ## Install
 
+Recommended for CLI isolation:
+
+```bash
+uv tool install young-stock-cli
+```
+
+Or install into the active Python environment:
+
 ```bash
 python3 -m pip install young-stock-cli
 ```
@@ -121,6 +129,61 @@ young config show       # secrets are masked
 young config path
 ```
 
+Model IDs are provider-specific and change over time. Do not rely on a console display name such as
+`Kimi-K2.6` or `ark-code-latest`; query the endpoint and copy an exact returned ID:
+
+```bash
+# Use saved provider/api_base/api_key_env settings
+young config models
+
+# Ark / Volcano Engine (OpenAI-compatible)
+export ARK_API_KEY="..."
+young config models \
+  --provider ark \
+  --api-key-env ARK_API_KEY
+
+# Kimi / Moonshot
+export MOONSHOT_API_KEY="..."
+young config models \
+  --provider kimi \
+  --api-key-env MOONSHOT_API_KEY
+
+# DeepSeek and Qwen can use their built-in API bases
+young config models --provider deepseek --api-key-env DEEPSEEK_API_KEY
+young config models --provider qwen --api-key-env DASHSCOPE_API_KEY
+
+# Local Ollama
+young config models --provider ollama --api-base http://localhost:11434/v1
+```
+
+For endpoint-level troubleshooting, the equivalent OpenAI-compatible request is:
+
+```bash
+curl -sS "$API_BASE/models" \
+  -H "Authorization: Bearer $MODEL_API_KEY" |
+python3 -c 'import json,sys; print("\n".join(x["id"] for x in json.load(sys.stdin).get("data", [])))'
+
+# Ark exact example
+curl -sS https://ark.cn-beijing.volces.com/api/v3/models \
+  -H "Authorization: Bearer $ARK_API_KEY" |
+python3 -c 'import json,sys; print("\n".join(x["id"] for x in json.load(sys.stdin).get("data", [])))'
+
+# Kimi / Moonshot exact example
+curl -sS https://api.moonshot.cn/v1/models \
+  -H "Authorization: Bearer $MOONSHOT_API_KEY" |
+python3 -c 'import json,sys; print("\n".join(x["id"] for x in json.load(sys.stdin).get("data", [])))'
+```
+
+For Ark, set `API_BASE=https://ark.cn-beijing.volces.com/api/v3`; for Kimi/Moonshot, set
+`API_BASE=https://api.moonshot.cn/v1`. After selecting an ID, pass the same API base to `young config llm`.
+Some coding-plan names and web-console aliases are not Chat Completions model IDs.
+
+```bash
+# Configure the exact ID returned above
+young config llm --provider ark --model "<ark-model-or-endpoint-id>" --api-key-env ARK_API_KEY
+young config llm --provider kimi --model "<moonshot-model-id>" --api-key-env MOONSHOT_API_KEY
+```
+
 Start the interactive mode with `young chat`. Slash commands reuse the same Click command tree, so traditional CLI
 and chat behavior stay aligned:
 
@@ -156,6 +219,11 @@ The report follows the six-module method from
 5. M5 market/portfolio style
 6. M6 resilient directions
 
+Before each LLM replay, young checks the remote `stock-analysis` `SKILL.md`. A newer text specification is cached
+under `~/.young_stock/methodologies/stock-analysis/` and used for report structure and writing discipline. Remote
+code is never executed; if the check is unavailable, the most recent cached specification or the bundled 4.2.0
+guidance is used.
+
 Each module has an evidence score. Missing fields remain missing rather than being rendered as zero. Low-quality
 evidence automatically produces a shorter report limited to verified indices, holdings, risks, and next-session
 checks. Markdown, metadata, and `evidence.json` are retained under:
@@ -168,6 +236,12 @@ checks. Markdown, metadata, and `evidence.json` are retained under:
 
 Install the optional renderer:
 
+```bash
+uv tool install --force 'young-stock-cli[pdf]'
+```
+
+If `young` was installed into the active Python environment instead:
+
 ```bash
 python3 -m pip install "young-stock-cli[pdf]"
 ```

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "young-stock-cli"
-version = "0.2.0"
+version = "0.2.1"
 description = "A-share (China stock market) after-hours CLI — no login, no scraping tricks, just data."
 readme = "README.md"
 requires-python = ">=3.9"

src/young_stock/__init__.py

@@ -1,3 +1,3 @@
 """young-stock-cli: A-share after-hours CLI."""
 
-__version__ = "0.2.0"
+__version__ = "0.2.1"

src/young_stock/_core.py

@@ -2169,6 +2169,17 @@ def fetch_eastmoney_board_list(board_type: str, date_str: str, limit: int = 100)
     return result
 
 
+def get_board_list(board_type: str, date_str: str, limit: int = 100) -> dict[str, Any]:
+    """Return board rankings through the stock-analysis source order."""
+    result = fetch_eastmoney_board_list(board_type, date_str, limit=limit)
+    if result.get("rows"):
+        return result
+    browser_result = camofox_board_list(board_type)
+    if browser_result.get("rows"):
+        return browser_result
+    return result
+
+
 @retry_on_recoverable(max_retries=MAX_RETRIES, initial_delay=INITIAL_BACKOFF)
 def eastmoney_datacenter(
     report_name: str,
@@ -2709,6 +2720,37 @@ def rank_symbols_by_news_heat(
 # 板块榜（camofox 降级层）
 # ------------------------------------------------------------------
 
+def _parse_camofox_board_snapshot(markdown: str) -> list[dict[str, Any]]:
+    rows = []
+    for line in markdown.splitlines():
+        line = line.strip()
+        if not line.startswith('row "'):
+            continue
+        content = line[5:].rstrip('"')
+        parts = re.split(r"\s{2,}", content)
+        if len(parts) < 4:
+            parts = content.split()
+        if len(parts) < 3:
+            continue
+        rank = _safe_int(parts[0])
+        if rank is None:
+            continue
+        rows.append(
+            {
+                "rank": rank,
+                "name": parts[1],
+                "change_pct": _safe_float(str(parts[2]).rstrip("%")),
+                "up_count": _safe_int(parts[3]) if len(parts) > 3 else None,
+                "down_count": _safe_int(parts[4]) if len(parts) > 4 else None,
+                "leader": parts[5] if len(parts) > 5 else "",
+                "leader_change_pct": (
+                    _safe_float(str(parts[6]).rstrip("%")) if len(parts) > 6 else None
+                ),
+            }
+        )
+    return rows
+
+
 def camofox_board_list(board_type: str = "industry") -> dict[str, Any]:
     base = os.environ.get("CAMOFOX_URL", "http://localhost:9377")
     user_id = os.environ.get("CAMOFOX_USER_ID", "")
@@ -2741,14 +2783,7 @@ def camofox_board_list(board_type: str = "industry") -> dict[str, Any]:
         with urllib.request.urlopen(urllib.request.Request(snap_url, method="GET"), timeout=15) as resp:
             md = resp.read().decode("utf-8", errors="ignore")
 
-        rows = []
-        for line in md.splitlines():
-            line = line.strip()
-            if line.startswith('row "'):
-                content = line[5:].rstrip('"')
-                parts = re.split(r"\s{2,}", content)
-                if len(parts) >= 4:
-                    rows.append(parts)
+        rows = _parse_camofox_board_snapshot(md)
         return {"board_type": board_type, "rows": rows, "count": len(rows)}
     except Exception as e:
         diag(f"camofox board {board_type}: {e}")
@@ -3927,12 +3962,8 @@ def run_a_share(date_str: str, include_news: bool = True) -> None:
     if include_news:
         print_a_share_news(date_str, zt)
 
-    industry = fetch_eastmoney_board_list("industry", date_str)
-    if not industry.get("rows"):
-        industry = camofox_board_list("industry")
-    concept = fetch_eastmoney_board_list("concept", date_str)
-    if not concept.get("rows"):
-        concept = camofox_board_list("concept")
+    industry = get_board_list("industry", date_str)
+    concept = get_board_list("concept", date_str)
     print_boards(industry, "行业板块涨幅")
     print_boards(concept, "概念板块涨幅")
 

src/young_stock/cli.py

@@ -22,6 +22,7 @@
 from .evidence import build_daily_evidence, build_stock_evidence
 from .llm import LLMClient, LLMError
 from .local_store import load_store, now_label, save_store, young_home
+from .methodology import sync_stock_analysis_methodology
 from .profile import (
     add_group,
     add_group_item,
@@ -125,7 +126,9 @@ def update(pre: bool, user_install: bool) -> None:
             "update failed with exit code "
             f"{result.returncode}. young-stock-cli requires Python 3.9+. "
             "Check `python3 --version`, then retry with "
-            "`python3 -m pip install --upgrade young-stock-cli`."
+            "`python3 -m pip install --upgrade young-stock-cli`；"
+            "如果 `which young` 指向 uv tool 环境，请运行 "
+            "`uv tool install --upgrade young-stock-cli`。"
         )
 
 
@@ -278,10 +281,17 @@ def _run_llm_replay(date_str: str, kind: str = "replay", symbol: str | None = No
     artifacts = ReportArtifacts(date_str)
     artifacts.write_json("evidence" if not symbol else f"{symbol}-evidence", evidence.to_dict())
     config = load_config(strict=False).get("llm", {})
+    methodology = sync_stock_analysis_methodology()
     try:
-        markdown, metadata = generate_llm_daily_report(evidence.to_dict(), LLMClient(config))
+        markdown, metadata = generate_llm_daily_report(
+            evidence.to_dict(),
+            LLMClient(config),
+            methodology=methodology.text,
+        )
     except LLMError as exc:
         raise click.ClickException(str(exc)) from exc
+    metadata["stock_analysis_version"] = methodology.version
+    metadata["stock_analysis_updated"] = methodology.updated
     name = f"analyze-{symbol}" if symbol else kind
     path = artifacts.write_markdown(name, markdown)
     artifacts.write_metadata({"kind": name, **metadata})
@@ -334,7 +344,11 @@ def config_show() -> None:
 
 
 @config.command("llm", help="Configure the LLM provider and model.")
-@click.option("--provider", required=True, type=click.Choice(["openai", "deepseek", "qwen", "ollama", "anthropic"]))
+@click.option(
+    "--provider",
+    required=True,
+    type=click.Choice(["openai", "ark", "kimi", "moonshot", "deepseek", "qwen", "ollama", "anthropic"]),
+)
 @click.option("--model", required=True)
 @click.option("--api-key", default=None, hide_input=True)
 @click.option("--api-key-env", default=None, help="Environment variable containing the API key.")
@@ -362,6 +376,43 @@ def config_llm(
     click.echo(f"LLM configured: provider={provider}; model={model}; config={config_path()}")
 
 
+@config.command("models", help="List model IDs exposed by an OpenAI-compatible, Anthropic, or Ollama endpoint.")
+@click.option(
+    "--provider",
+    default=None,
+    type=click.Choice(["openai", "ark", "kimi", "moonshot", "deepseek", "qwen", "ollama", "anthropic"]),
+)
+@click.option("--api-key", default=None, hide_input=True)
+@click.option("--api-key-env", default=None, help="Environment variable containing the API key.")
+@click.option("--api-base", default=None, help="Provider API base, for example https://api.moonshot.cn/v1.")
+@click.option("--timeout", default=30, show_default=True, type=float)
+def config_models(
+    provider: str | None,
+    api_key: str | None,
+    api_key_env: str | None,
+    api_base: str | None,
+    timeout: float,
+) -> None:
+    saved = dict(load_config(strict=False).get("llm") or {})
+    query = {
+        **saved,
+        "provider": provider or saved.get("provider"),
+        "api_key": api_key if api_key is not None else saved.get("api_key"),
+        "api_key_env": api_key_env if api_key_env is not None else saved.get("api_key_env"),
+        "api_base": api_base or saved.get("api_base"),
+        "timeout": timeout,
+    }
+    try:
+        models = LLMClient(query).list_models()
+    except LLMError as exc:
+        raise click.ClickException(str(exc)) from exc
+    if not models:
+        click.echo("该服务当前未返回可用模型 ID。")
+        return
+    for model_id in models:
+        click.echo(model_id)
+
+
 @config.group("channel", help="Manage notification channels.")
 def config_channel() -> None:
     pass

src/young_stock/evidence.py

@@ -62,6 +62,26 @@ def _pool(data: dict[str, Any]) -> tuple[int | None, list[dict[str, Any]]]:
     return payload.get("tc", len(rows)), rows
 
 
+def _board_list(core: Any, board_type: str, trade_date: str) -> dict[str, Any]:
+    if hasattr(core, "get_board_list"):
+        return _call({}, core.get_board_list, board_type, trade_date, limit=100)
+    result = _call({}, core.fetch_eastmoney_board_list, board_type, trade_date, limit=100)
+    if result.get("rows") or not hasattr(core, "camofox_board_list"):
+        return result
+    browser_result = _call({}, core.camofox_board_list, board_type)
+    return browser_result if browser_result.get("rows") else result
+
+
+def _fund_flow_available(data: dict[str, Any]) -> bool:
+    if not isinstance(data, dict) or data.get("_unavailable") or data.get("_error"):
+        return False
+    return any(
+        isinstance(value, (int, float)) or bool(value)
+        for key, value in data.items()
+        if not str(key).startswith("_") and key != "date"
+    )
+
+
 def build_daily_evidence(core: Any, trade_date: str, profile: dict[str, Any] | None = None) -> EvidenceBundle:
     profile = profile or {}
     a_indices = [_index_dict(item) for item in _call([], core.get_index, trade_date)]
@@ -78,9 +98,9 @@ def build_daily_evidence(core: Any, trade_date: str, profile: dict[str, Any] | N
         trade_date,
     )
     northbound = _call({}, core.fetch_northbound_flow_snapshot, trade_date)
-    fund_flow = _call({}, core.get_fund_flow, trade_date, strict_date=False)
-    industry = _call({}, core.fetch_eastmoney_board_list, "industry", trade_date, limit=100)
-    concept = _call({}, core.fetch_eastmoney_board_list, "concept", trade_date, limit=100)
+    fund_flow = _call({}, core.get_fund_flow, trade_date, strict_date=True)
+    industry = _board_list(core, "industry", trade_date)
+    concept = _board_list(core, "concept", trade_date)
     zt = _call({}, core.get_zt_pool, trade_date)
     dt = _call({}, core.get_dt_pool, trade_date)
     zb = _call({}, core.get_zb_pool, trade_date)
@@ -113,7 +133,7 @@ def build_daily_evidence(core: Any, trade_date: str, profile: dict[str, Any] | N
             },
         },
         "M2": {
-            "available": bool(board_rows),
+            "available": bool(board_rows) or _fund_flow_available(fund_flow),
             "industry": industry.get("rows") or [],
             "concept": concept.get("rows") or [],
             "fund_flow": fund_flow,