Merge branch 'codex/llm-report-data-fixes'

Author: yjw

CHANGELOG.md

@@ -5,17 +5,39 @@ 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).
 
+## [0.2.11] - 2026-06-20
+
+### Changed
+
+- Removed the deprecated `young replay` compatibility command so `young daily --llm` is now the only LLM deep-replay entry.
+- Tightened the chat slash-command surface by removing the old `/replay` and `/daily-llm` compatibility aliases.
+- Clarified throughout the CLI help and README that plain `young daily` is deterministic and does not require any LLM configuration.
+- Updated `young send` docs to match the current delivery contract: send the latest Markdown plus summary first, and attach the same-name PDF only when it exists.
+
+### Fixed
+
+- Added regression coverage to keep deterministic `young daily` on the non-LLM path unless `--llm` is explicitly requested.
+- Added regression coverage for latest-Markdown selection when `young send` is run without an explicit date.
+
+## [0.2.10] - 2026-06-19
+
+### Fixed
+
+- Hardened persisted LLM auth so saved credentials survive terminal restarts even when a stale environment variable is present.
+- Switched `young chat` input handling to a prompt toolkit-backed editor to reduce backspace/cursor glitches on macOS terminals.
+
 ## [0.2.9] - 2026-06-19
 
 ### Changed
 - `young chat` now auto-uses the existing `reach` bridge for explicit finance/news lookup requests, but only feeds compact evidence excerpts to the model instead of exposing raw search captures in the terminal.
-- `young chat` input now uses a fixed `young ` prompt prefix so line editing no longer eats the visible prompt label.
+- `young chat` input now uses `prompt_toolkit` with a fixed, non-editable `young ` prompt prefix for reliable backspace and cursor movement on macOS terminals.
 - Help text now makes the command boundary clearer: `young daily --llm` is the strict stock-analysis M1-M6 Markdown replay, `young replay` is only a deprecated alias, and `young report` is PDF export only.
 
 ### Fixed
 - Fixed chat search flows that previously surfaced raw `/reach` output or still nudged users to run `/reach` manually instead of returning a summarized answer directly.
 - Hardened LLM replay prompts so `young daily --llm` stays on the stock-analysis six-module framework and does not drift into persona-style investment templates.
 - Fixed LLM config persistence so `young config llm --api-key-env ...` also stores a local fallback key, which keeps `young chat` working across fresh terminal sessions without forcing users to re-export the secret every time.
+- Fixed stale shell environment variables overriding the saved API key after terminal restarts; saved credentials now take precedence and accidental outer quotes or whitespace are normalized.
 
 ## [0.2.8] - 2026-06-19
 

README.md

@@ -84,18 +84,17 @@ young profile add-fund 021528 --buy-date 2026-01-10 --quantity 1000
 young profile list
 young profile clear-stocks  # clear all saved stocks/ETFs only
 young profile clear-funds   # clear all saved funds only
-young daily --format summary      # deterministic watchlist daily report
-young daily --format key-points   # short deterministic report with trend/risk points
-young daily --format full         # full deterministic watchlist daily report
-young daily --llm                # deep after-hours / latest-trading-day replay; saves Markdown
+young daily --format summary      # deterministic watchlist daily report; no LLM needed
+young daily --format key-points   # short deterministic report with trend/risk points; no LLM needed
+young daily --format full         # full deterministic watchlist daily report; no LLM needed
+young daily --llm                 # the only deep after-hours / latest-trading-day replay entry; saves Markdown
 young daily --llm --refresh       # rebuild the same identity from scratch
 young init                       # initialize local state and verify report/PDF readiness
-young replay                     # deprecated alias for young daily --llm
 young analyze 600519             # deep single-stock analysis
 young reach 贵州茅台 盈利 新闻    # optional Agent-Reach bridge for external web/company research
 young chat                       # Rich chat mode with slash commands
 young report                     # export the latest Markdown report to PDF
-young send                       # send latest Markdown + PDF to configured channels
+young send                       # send latest Markdown + summary; attach same-name PDF only if it exists
 young daily --only 基金,A股 --quick
 young news 3690.HK      # multi-source news only
 young diagnose          # network/source diagnostic
@@ -206,8 +205,6 @@ commands through Click:
 /reach 贵州茅台 盈利 新闻
 /daily --format summary
 /daily --llm
-/daily-llm
-/replay
 /profile list
 /style
 /style list
@@ -218,10 +215,10 @@ commands through Click:
 ```
 
 The chat keeps the five most recent user/assistant turns. If no LLM is configured, enhanced commands fail with a
-configuration hint while `/a`, `/stock`, `/daily`, and every other deterministic command continue working. Chat is
-authoritative and only exposes a white-listed command surface: do not expect `/market` or `/trend`, and `/send`,
-`/config`, `/update`, and `/uninstall` are intentionally blocked from chat. `/daily-llm` and `/replay` are kept as
-deprecated aliases that route to `/daily --llm`.
+configuration hint while `/a`, `/stock`, `/daily`, and every other deterministic command continue working. `/daily`
+without `--llm` stays deterministic and does not require LLM. Chat is authoritative and only exposes a white-listed
+command surface: do not expect `/market` or `/trend`, and `/send`, `/config`, `/update`, and `/uninstall` are
+intentionally blocked from chat.
 
 The style commands persist under chat config and synchronously set dialogue tone, self-reference style, and analysis
 framework. Supported styles are `balanced`, `buffett`, `munger`, `graham`, and `dalio`.
@@ -233,9 +230,8 @@ without making the main CLI package heavy.
 
 ### Evidence-driven deep replay
 
-`young daily --llm` builds an Evidence Pack entirely from the structured values returned by `young_stock._core`.
-`young replay` and `/replay` are deprecated aliases for the same workflow. The model is used for synthesis, not data
-collection.
+`young daily --llm` builds an Evidence Pack entirely from the structured values returned by `young_stock._core`. It is
+the only LLM replay entry. The model is used for synthesis, not data collection.
 
 The report follows the six-module method from
 [`AdvancingTitans/stock-analysis`](https://github.com/AdvancingTitans/stock-analysis):
@@ -329,7 +325,7 @@ young config channel add feishu work \
   --webhook "https://open.feishu.cn/open-apis/bot/v2/hook/..."
 ```
 
-App mode uploads and sends both Markdown and PDF:
+App mode uploads and sends Markdown first, then adds the same-name PDF when it exists:
 
 ```bash
 young config channel add feishu research-app \
@@ -384,8 +380,8 @@ The internals are being split into focused modules: `young_stock.calendar` handl
 - **Multiple public quote sources** — Sina Finance and Tencent Finance are the default stable quote path; market-specific public interfaces stay behind the CLI fallback layer when no better source is available.
 - **Single-stock lookup** — `young stock 600519`, `young stock 0700.HK`, or `young stock AAPL` prints a compact quote snapshot with source, trade date, price, change, volume, turnover, market cap, PE/PB, 52-week range when available, and optional news.
 - **Fund holding lookup** — `young fund 161725` prints the fund's same-day estimated change, latest NAV date, top holdings, holding-stock quotes, rough contribution estimate, and same-day holding-stock news. Official fund NAVs usually update at night, so intraday/close values are clearly labeled as estimates.
-- **Personal daily report** — `young daily` is the deterministic watchlist report from your local investment memory in `~/.young_stock/profile.json`; it prints saved stock/ETF trends, fund estimates, only the markets relevant to your symbols and fund top holdings, and portfolio-style suggestions grounded in your funds, stocks, holding dates, quantities, and available news. First use requires a verified symbol plus `--buy-date` and `--quantity`, for example `young profile add-stock 600519 --buy-date 2026-01-15 --quantity 100` or `young profile add-fund 161725 --buy-date 2026-01-10 --quantity 1000`.
-- **LLM replay** — `young daily --llm` performs the deep after-hours / latest-trading-day replay, writes the matching Markdown artifact under the same report identity, and prints it in the terminal; `young report` is the dedicated PDF export step, and `--refresh` rebuilds the Markdown identity. `young replay`, `/replay`, and `/daily-llm` are compatibility aliases only.
+- **Personal daily report** — `young daily` is the deterministic watchlist report from your local investment memory in `~/.young_stock/profile.json`; it does not require LLM, and it prints saved stock/ETF trends, fund estimates, only the markets relevant to your symbols and fund top holdings, and portfolio-style suggestions grounded in your funds, stocks, holding dates, quantities, and available news. First use requires a verified symbol plus `--buy-date` and `--quantity`, for example `young profile add-stock 600519 --buy-date 2026-01-15 --quantity 100` or `young profile add-fund 161725 --buy-date 2026-01-10 --quantity 1000`.
+- **LLM replay** — `young daily --llm` performs the deep after-hours / latest-trading-day replay, writes the matching Markdown artifact under the same report identity, and prints it in the terminal; `young report` is the dedicated PDF export step, and `--refresh` rebuilds the Markdown identity.
 - **Short report modes** — `young daily --format summary` keeps terminal output compact; `--format key-points` adds a few trend/risk bullets; `--only`, `--order`, and `--quick` trim slower or irrelevant sections.
 - **Chat style control** — `/style list|set|show|clear` persists one synchronized style choice for tone, self-reference, and analysis framework. `balanced`, `buffett`, `munger`, `graham`, and `dalio` are style contracts, not identity claims.
 - **Investment memory management** — list, remove, clear, and group saved stocks/funds with `young profile list`, `remove-stock`, `remove-fund`, `clear`, `clear-stocks`, `clear-funds`, and `profile group create/add`.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "young-stock-cli"
-version = "0.2.9"
+version = "0.2.11"
 description = "A-share (China stock market) after-hours CLI — no login, no scraping tricks, just data."
 readme = "README.md"
 requires-python = ">=3.9"
@@ -29,6 +29,7 @@ dependencies = [
     "requests>=2.31",
     "rich>=13.0",
     "click>=8.1",
+    "prompt-toolkit>=3.0,<4",
     "weasyprint>=62",
 ]
 

src/young_stock/__init__.py

@@ -1,3 +1,3 @@
 """young-stock-cli: A-share after-hours CLI."""
 
-__version__ = "0.2.9"
+__version__ = "0.2.11"

src/young_stock/channels/__init__.py

@@ -11,12 +11,13 @@
 
 
 def send_report(trade_date: str | None, *, channel_name: str | None = None) -> list[DeliveryResult]:
-    if not trade_date:
-        raise ValueError("没有报告日期；请先运行 `young report`。")
     markdown = ReportArtifacts.latest_markdown(trade_date)
-    pdf = markdown.with_suffix(".pdf") if markdown else None
-    if markdown is None or not pdf.exists():
-        raise ValueError(f"{trade_date} 缺少 Markdown/PDF；请先运行 `young report --date {trade_date}`。")
+    if markdown is None:
+        if trade_date:
+            raise ValueError(f"{trade_date} 缺少可发送的 Markdown；请先生成对应日报或 LLM Markdown。")
+        raise ValueError("未找到可发送的 Markdown；请先生成日报或 LLM Markdown。")
+    pdf_candidate = markdown.with_suffix(".pdf")
+    pdf = pdf_candidate if pdf_candidate.exists() else None
     configs = load_config(strict=False).get("channels", {}).get("feishu", {})
     selected = {
         name: config

src/young_stock/channels/feishu.py

@@ -40,9 +40,15 @@ def _post(self, url: str, **kwargs: Any) -> dict[str, Any]:
             return data
         raise RuntimeError("Feishu request retries exhausted")
 
-    def _preview(self, markdown: Path, pdf: Path, token: str | None = None) -> None:
+    def _preview(self, markdown: Path, pdf: Path | None, token: str | None = None) -> None:
         content = markdown.read_text(encoding="utf-8")[:3000]
-        text = f"{markdown.stem}\n\n{content}\n\nPDF 本地路径: {pdf}"
+        if pdf is None:
+            delivery_note = "未检测到同名 PDF，本次仅发送 Markdown 摘要。"
+        elif token:
+            delivery_note = "同名 PDF 已检测到，将随后的文件消息一并发送。"
+        else:
+            delivery_note = "同名 PDF 已检测到，但 webhook 不支持附件上传。"
+        text = f"{markdown.stem}\n\n{content}\n\n{delivery_note}"
         if token:
             self._send_message("text", {"text": text}, token)
         else:
@@ -89,15 +95,23 @@ def _send_message(self, msg_type: str, content: dict[str, Any], token: str) -> N
             },
         )
 
-    def send(self, markdown: Path, pdf: Path) -> DeliveryResult:
+    def send(self, markdown: Path, pdf: Path | None) -> DeliveryResult:
         try:
             if self.config.get("webhook"):
                 self._preview(markdown, pdf)
-                return DeliveryResult("feishu", self.name, True, "已发送 Markdown 预览；webhook 不支持附件上传。")
+                if pdf is None:
+                    detail = "已发送 Markdown 摘要；未检测到同名 PDF。"
+                else:
+                    detail = "已发送 Markdown 摘要；检测到同名 PDF，但 webhook 不支持附件上传。"
+                return DeliveryResult("feishu", self.name, True, detail)
             token = self._token()
             self._preview(markdown, pdf, token)
-            for path in (markdown, pdf):
-                self._send_message("file", {"file_key": self._upload(path, token)}, token)
-            return DeliveryResult("feishu", self.name, True, "Markdown 与 PDF 已上传并发送。")
+            self._send_message("file", {"file_key": self._upload(markdown, token)}, token)
+            if pdf is not None:
+                self._send_message("file", {"file_key": self._upload(pdf, token)}, token)
+                detail = "Markdown 与 PDF 已上传并发送。"
+            else:
+                detail = "Markdown 已上传并发送；未检测到同名 PDF。"
+            return DeliveryResult("feishu", self.name, True, detail)
         except Exception as exc:
             return DeliveryResult("feishu", self.name, False, str(exc))

src/young_stock/chat.py

@@ -17,10 +17,10 @@
 from rich.console import Console
 from rich.markdown import Markdown
 
-try:  # ponytail: best-effort line editing on local terminals; falls back harmlessly if unavailable.
-    import readline  # noqa: F401
-except Exception:  # pragma: no cover - platform dependent
-    readline = None
+try:  # pragma: no cover - optional dependency may be absent in some envs.
+    from prompt_toolkit import prompt as _prompt_toolkit_prompt
+except Exception:  # pragma: no cover - dependency missing or broken
+    _prompt_toolkit_prompt = None
 
 try:  # pragma: no cover - import path differs across Python builds
     from zoneinfo import ZoneInfo
@@ -297,6 +297,17 @@ def _current_time_system_note() -> str:
     )
 
 
+def _read_chat_input(prompt_text: str = "young ") -> str:
+    if callable(_prompt_toolkit_prompt):
+        try:
+            return _prompt_toolkit_prompt(prompt_text)
+        except (EOFError, KeyboardInterrupt):
+            raise
+        except Exception:
+            pass
+    return builtins.input(prompt_text)
+
+
 def _is_time_query(text: str) -> bool:
     stripped = text.strip()
     if not stripped or stripped.startswith("/"):
@@ -618,7 +629,7 @@ def _build_messages(self, extra_system_messages: list[str] | None = None) -> lis
                     "4) 用户要做单只股票深入分析、但你缺少足够证据时，不要直接拒绝；"
                     "优先使用已有的 /stock <symbol>、/analyze <symbol> 或已注入的外部检索证据继续分析。"
                     "5) 如果本轮已经完成外部检索，就直接总结结果，不要要求用户自己执行 /reach。"
-                    "6) /daily --llm 与 /replay 的深度复盘必须严格遵循 stock-analysis 的 M1-M6 框架，不得改写成巴菲特、芒格等个体投资框架。"
+                    "6) /daily --llm 的深度复盘必须严格遵循 stock-analysis 的 M1-M6 框架，不得改写成巴菲特、芒格等个体投资框架。"
                     "7) 风格与长期记忆都服从本安全规则。"
                 ),
             }
@@ -685,8 +696,6 @@ def _resolve_click_args(self, args: list[str]) -> tuple[list[str] | None, str |
             if len(args) >= 2 and args[1] == "clear":
                 return args, None
             return None, "chat 中仅支持 /memory show 和 /memory clear。"
-        if root in {"daily-llm", "replay"}:
-            return ["daily", "--llm", *args[1:]], "提示：/daily-llm 和 /replay 已弃用，请改用 /daily --llm。"
         if root in {"send", "config", "update", "uninstall"}:
             return None, f"chat 中禁止 /{root}，请在终端直接运行对应 CLI。"
         return None, f"不支持 /{root}。请输入 /help 查看 authoritative 命令列表。"
@@ -788,7 +797,7 @@ def run_chat() -> None:
     )
     while True:
         try:
-            text = builtins.input("young ").strip()
+            text = _read_chat_input().strip()
         except (EOFError, KeyboardInterrupt):
             console.print("\n再见。")
             return