@jayyuen66/dsh-ocr-review 0.1.9 → 0.1.10

package/docs/DETAIL.md

@@ -4,10 +4,10 @@ Runs the external `ocr` command line over a change set, reports the comments it
 
 ## What the card changes
 
-- **Model** — `ocrProvider` and `ocrModel` are not stored settings: they project the host's model namespace, and entries without a usable key are shown but disabled. `chatGptModel` reflects the external CLI's own login state, which this plugin reads but does not own.
-- **Review parameters** — `effort` (`auto`/`low`/`medium`/`high`) is how many review passes run: deeper review costs more time. `language` chooses Chinese or English output. `autoVerify` (on) runs `ocr llm test` after you pick a provider and model, so a wrong choice fails at pick time rather than at first review. `maxComments` (default 12; 0 means all, including comments that cannot be anchored to a line). `timeoutMinutes` (0 requests the host's ceiling; a positive value is minutes, still clamped by the host). `ocrConfigPath` — empty means `~/.opencodereview/config.json`; a non-empty value must be `<dir>/.opencodereview/config.json` and is honoured by redirecting `HOME` (POSIX only).
+- **Model** — `ocrProvider` and `ocrModel` are not stored settings: they project the host's model namespace, and entries without a usable key are shown but disabled. `chatGptModel` is a slug this plugin writes into the CLI's own config (provider `chatgpt`, with `api_key` / `api_key_cmd` removed); the credentials behind that provider come from the CLI's own login state, which the plugin only queries (`ocr auth`) and never stores.
+- **Review parameters** — `effort` (`auto`/`low`/`medium`/`high`) is how many review passes run: deeper review costs more time. `language` chooses Chinese or English output. `autoVerify` (on) runs `ocr llm test` after you pick a provider and model, so a wrong choice fails at pick time rather than at first review. `maxComments` (default 12; 0 means all, including comments that cannot be anchored to a line). `timeoutMinutes` (0 requests the host's ceiling; a positive value is minutes, still clamped by the host). `ocrConfigPath` — empty means `~/.opencodereview/config.json`; a non-empty value must be `<dir>/.opencodereview/config.json` and is honoured by redirecting `HOME`; on Windows the value passes the shape check but the redirection does not happen, and the CLI keeps its default path.
 
-Six of these are read per call, so a change applies on the next review without restarting the host.
+`effort`, `maxComments`, `timeoutMinutes` and `ocrConfigPath` are read per call, so a change applies on the next review without restarting the host. `language` and `autoVerify` are read when a provider/model selection is applied: `language` is written into the CLI config at that moment, and `autoVerify` only decides whether `ocr llm test` runs then.
 
 ## Two separate things called MCP
 
@@ -19,7 +19,7 @@ The card's MCP-servers editor writes the **external `ocr` tool's own** configura
 
 ## What it needs from outside
 
-The `ocr` command line, expected at version **1.13.3**; the expectation is declared in code and mirrored in the package's manifest and READMEs, so a mismatch is reported rather than papered over. Install paths are `brew install open-code-review` or `npm i -g @alibaba-group/open-code-review@1.13.3`. A missing binary is reported as "command not found"; the case where a launcher exists but the binary does not says exactly that. The version probe always forces `OCR_NO_UPDATE=1` so probing cannot trigger an upgrade.
+The `ocr` command line, expected at version **1.13.3**; the expectation is declared once as `ocrCliVersion` in `package.json`, and the code constant plus both READMEs mirror that string, so a mismatch is reported (as a hint, never a hard failure) rather than papered over. Install paths are `brew install open-code-review` or `npm i -g @alibaba-group/open-code-review@1.13.3`. With no candidate on PATH at all, the pre-check reports "no ocr candidate on either path" plus an install hint; "command not found" (exit 127) is the different case where a candidate exists but the binary cannot actually be run. The version probe always forces `OCR_NO_UPDATE=1` so probing cannot trigger an upgrade.
 
 API credentials are not stored here: the plugin writes a `api_key_cmd` that calls a helper script, so the key is fetched at run time instead of landing in a config file. If that script is missing, the card warns rather than sending an unauthenticated request.
 

package/docs/DETAIL.zh.md

@@ -4,10 +4,10 @@
 
 ## 设置卡上改的是什么
 
-- **模型** — `ocrProvider` 与 `ocrModel` 不是本插件存的值：它们是宿主模型命名空间的投影，没有可用密钥的条目照样列出但禁用。`chatGptModel` 反映的是外部命令行自己的登录态，本插件读它但不拥有它。
-- **审查参数** — `effort`（`auto`/`low`/`medium`/`high`）是跑几轮审查，越深越费时。`language` 选中文或英文输出。`autoVerify`（默认开）在选完提供方与模型后跑一次 `ocr llm test`，选错当场就知道，而不是等第一次评审失败。`maxComments`（默认 12；0 表示全部，包括钉不到行上的意见）。`timeoutMinutes`（0 表示向宿主要上限；正数是分钟数，仍由宿主收口）。`ocrConfigPath`——留空就是 `~/.opencodereview/config.json`；填了就必须是 `<目录>/.opencodereview/config.json` 的形状，生效方式是重定向 `HOME`（仅 POSIX）。
+- **模型** — `ocrProvider` 与 `ocrModel` 不是本插件存的值：它们是宿主模型命名空间的投影，没有可用密钥的条目照样列出但禁用。`chatGptModel` 是本插件写进外部命令行自己配置的 model slug（provider 置为 `chatgpt`，同时删掉 `api_key` / `api_key_cmd`）；该提供方背后的凭据来自命令行自己的登录态，本插件只查询 `ocr auth`，不保存。
+- **审查参数** — `effort`（`auto`/`low`/`medium`/`high`）是跑几轮审查，越深越费时。`language` 选中文或英文输出。`autoVerify`（默认开）在选完提供方与模型后跑一次 `ocr llm test`，选错当场就知道，而不是等第一次评审失败。`maxComments`（默认 12；0 表示全部，包括钉不到行上的意见）。`timeoutMinutes`（0 表示向宿主要上限；正数是分钟数，仍由宿主收口）。`ocrConfigPath`——留空就是 `~/.opencodereview/config.json`；填了就必须是 `<目录>/.opencodereview/config.json` 的形状，生效方式是重定向 `HOME`；Windows 上值能通过形状校验但不会重定向，命令行仍读默认路径。
 
-其中六项每次调用现读，改完下一次评审生效，不需要重启宿主。
+`effort`、`maxComments`、`timeoutMinutes` 与 `ocrConfigPath` 每次调用现读，改完下一次评审生效，不需要重启宿主。`language` 与 `autoVerify` 在应用一次 provider/model 选择时读取：`language` 当场写进命令行配置，`autoVerify` 只决定那时是否跑 `ocr llm test`。
 
 ## 两处都叫 MCP，但不是同一件事
 
@@ -15,11 +15,11 @@
 
 ## 要重启的值
 
-`llmTestTimeoutMs`（90000）、`startupVerify`（默认开：装载时自检端点）、`ocrBackgroundMaxMs`（0 表示不设兜底窗口）、`stdoutMaxBytes`（400000）、`ocrNoAutoUpdate`（默认 false，即保留上游的自动升级，除非你明确关掉）、`ocrInstallHint`（`both`/`brew`/`npm`）。写在本插件注册行的 `config:` 块，改值随重启生效。
+`llmTestTimeoutMs`（90000）、`startupVerify`（默认开：装载时自检端点）、`ocrBackgroundMaxMs`（0 表示不设兜底窗口）、`stdoutMaxBytes`（400000）、`ocrNoAutoUpdate`（默认 false，即保留上游的自动升级，除非你明确关掉）、`ocrInstallHint`（`both`/`brew`/`npm`）。写在本插件于 `cordis.patch.yml` 注册行的 `config:` 块，改值随重启生效。
 
 ## 需要外部提供什么
 
-`ocr` 命令行，期望版本 **1.13.3**；这个期望写在代码里，并镜像到包清单与两份 README，所以版本不符是被报出来的，不是被掩盖的。装法是 `brew install open-code-review` 或 `npm i -g @alibaba-group/open-code-review@1.13.3`。命令行不存在时按「找不到命令」报；只有垫片没有真二进制时说的也是这一件。版本探测一律强制 `OCR_NO_UPDATE=1`，以免探测顺手触发升级。
+`ocr` 命令行，期望版本 **1.13.3**；这个期望唯一声明在 `package.json` 的 `ocrCliVersion`，代码常量与两份 README 镜像同一个字符串，版本不符只作为提示报出，从不中断评审。装法是 `brew install open-code-review` 或 `npm i -g @alibaba-group/open-code-review@1.13.3`。PATH 上一个候选都没有时，预检报「no ocr candidate on either path」并附安装提示；「command not found」（exit 127）是另一种情况：候选在，但真二进制跑不起来。版本探测一律强制 `OCR_NO_UPDATE=1`，以免探测顺手触发升级。
 
 API 凭据不存这里：本插件写的是一个 `api_key_cmd`，运行时的密钥由那支辅助脚本现取，不进配置文件。脚本缺失时卡上给出告警，而不是发一次没有鉴权的请求。