@praeviso/code-env-switch 0.1.6 → 0.1.7

package/README.md

@@ -67,6 +67,8 @@ npm link
 > `codenv init`, the shell wrapper applies them automatically.
 > The snippet also wraps `codex`/`claude` to bind sessions to profiles; use
 > `command codex` / `command claude` to bypass.
+> When you switch profiles within the same session, usage tracking and cost
+> display follow the most recently applied profile for subsequent tokens.
 
 ### Common commands
 
@@ -90,36 +92,6 @@ codenv usage-reset --yes
 
 This deletes usage history files (`usage.jsonl`, usage state, `profile-log.jsonl`, `statusline-debug.jsonl`) plus any backup variants in the config directory.
 
-### Usage tracking
-
-Usage stats are derived from two sources (statusline input + session logs) and appended to `usage.jsonl`.
-
-- Files and paths
-  - `usage.jsonl`: JSONL records with `ts`, `type`, `profileKey`/`profileName`, `model`, `sessionId`, and token breakdowns.
-  - `usage.jsonl.state.json`: per-session totals + per-session-file metadata (mtime/size) used to compute deltas and avoid double counting.
-  - `profile-log.jsonl`: profile usage + session binding log (`use` and `session` events).
-  - `statusline-debug.jsonl`: optional debug capture when `CODE_ENV_STATUSLINE_DEBUG` is enabled.
-  - Paths can be overridden via `usagePath`, `usageStatePath`, `profileLogPath`, `codexSessionsPath`, `claudeSessionsPath`.
-- Session binding (profile -> session)
-  - `codenv init` installs a shell wrapper so `codex`/`claude` run via `codenv launch`.
-  - `codenv launch` logs profile usage and then finds the latest unbound session file (prefers matching `cwd`, within a small grace window) and records the binding in `profile-log.jsonl`.
-  - Sync uses these bindings to attribute session usage to a profile.
-- Statusline sync (`codenv statusline --sync-usage`)
-  - Requires `sessionId`, model, and a profile (`profileKey` or `profileName`) to write usage.
-  - Reads stdin JSON for totals and computes a delta against the last stored totals in the state file.
-  - If totals decrease (session reset), the current totals are treated as fresh usage; otherwise negative sub-deltas are clamped to zero.
-  - If the token breakdown exceeds the reported total, the breakdown sum wins.
-  - Appends a delta record to `usage.jsonl` and updates the per-session totals in the state file.
-- Session log sync (`--sync-usage` and `codenv list`)
-  - Scans Codex sessions under `CODEX_HOME/sessions` (or `~/.codex/sessions`) and Claude sessions under `CLAUDE_HOME/projects` (or `~/.claude/projects`).
-  - Codex: parses `event_msg` token_count records and uses max totals; cached input is recorded as cache read and subtracted from input when possible.
-  - Claude: sums `message.usage` input/output/cache tokens across the file.
-  - Deltas are computed against prior file metadata and per-session maxima in the state file; files without a resolved binding are skipped.
-- Daily totals
-  - "Today" is computed in local time between 00:00 and the next 00:00.
-- Cost calculation
-  - Uses pricing from the profile (or `pricing.models`, plus defaults) and requires token splits; if splits are missing, cost is omitted.
-
 ### Add / update a profile
 
 ```bash

package/README_zh.md

@@ -67,6 +67,8 @@ npm link
 > shell 包装函数会自动在当前终端生效。
 > 该片段还会包装 `codex`/`claude` 以绑定会话到 profile；如需绕过，
 > 可使用 `command codex` / `command claude`。
+> 同一个 session 内切换 profile 时，后续用量与金额会跟随最新应用的
+> profile 统计。
 
 ### 常用命令
 
@@ -90,36 +92,6 @@ codenv usage-reset --yes
 
 该命令会删除用量历史文件（`usage.jsonl`、用量 state、`profile-log.jsonl`、`statusline-debug.jsonl`）以及配置目录中的相关备份文件。
 
-### 用量统计逻辑
-
-用量来自两条路径（状态栏输入同步 + 会话日志解析），最终追加到 `usage.jsonl`。
-
-- 文件与路径
-  - `usage.jsonl`：JSONL 记录，包含 `ts`/`type`/`profileKey`/`profileName`/`model`/`sessionId` 和 token 拆分字段。
-  - `usage.jsonl.state.json`：保存每个 session 的累计 totals + session 文件的 mtime/size，用于计算增量并避免重复统计。
-  - `profile-log.jsonl`：profile 使用与 session 绑定日志（`use`/`session`）。
-  - `statusline-debug.jsonl`：当 `CODE_ENV_STATUSLINE_DEBUG` 开启时写入的调试信息。
-  - 可通过 `usagePath`/`usageStatePath`/`profileLogPath`/`codexSessionsPath`/`claudeSessionsPath` 覆盖默认路径。
-- 会话绑定（profile -> session）
-  - `codenv init` 安装 shell 包装函数，使 `codex`/`claude` 实际走 `codenv launch`。
-  - `codenv launch` 记录 profile 使用，并在启动后短时间内（默认约 5 秒、每秒轮询）找到最新未绑定的 session 文件（优先 `cwd` 匹配），写入 `profile-log.jsonl`。
-  - 后续同步会用该绑定将 session 归因到对应 profile。
-- 状态栏同步（`codenv statusline --sync-usage`）
-  - 需要 `sessionId`、model，以及 profile（`profileKey` 或 `profileName`）才会写入用量。
-  - 从 stdin JSON 读取 totals，与 state 中的上次 totals 做差得到增量。
-  - 如果 totals 回退（session reset），直接把当前 totals 当作新增；否则对负数子项做 0 处理。
-  - 当拆分合计大于 total 时，以拆分合计为准。
-  - 把增量写入 `usage.jsonl`，并更新 state 中的 session totals。
-- 会话日志同步（`--sync-usage` 与 `codenv list`）
-  - 扫描 Codex 的 `CODEX_HOME/sessions`（或 `~/.codex/sessions`）与 Claude 的 `CLAUDE_HOME/projects`（或 `~/.claude/projects`）。
-  - Codex：读取 `event_msg` 的 token_count 记录，取累计最大值；cached input 计为 cache read，并在可判断时从 input 中扣除。
-  - Claude：汇总 `message.usage` 中的 input/output/cache tokens。
-  - 与 state 中的文件元数据和 session 最大值做差，生成增量并写入 `usage.jsonl`；无法解析到绑定的文件会被跳过。
-- 今日统计
-  - “今日”按本地时区 00:00 到次日 00:00 计算。
-- 费用换算
-  - 使用 profile 定价或 `pricing.models`（含默认值）换算，依赖 input/output/cache 拆分；缺少拆分则不显示金额。
-
 ### 添加 / 更新 profile
 
 ```bash

package/bin/usage/index.js

@@ -449,6 +449,8 @@ function syncUsageFromStatuslineInput(config, configPath, type, profileKey, prof
             endTs: now,
             cwd: cwd || (prev ? prev.cwd : null),
             model: resolvedModel,
+            profileKey: profileKey || null,
+            profileName: profileName || null,
         };
         state.sessions = sessions;
         updateUsageStateMetadata(state, usagePath);
@@ -1142,6 +1144,10 @@ function buildUsageSessionsFromRecords(records) {
         entry.cacheReadTokens += toUsageNumber(record.cacheReadTokens);
         entry.cacheWriteTokens += toUsageNumber(record.cacheWriteTokens);
         entry.totalTokens += toUsageNumber(record.totalTokens);
+        if (record.profileKey)
+            entry.profileKey = record.profileKey;
+        if (record.profileName)
+            entry.profileName = record.profileName;
         if (!entry.model && record.model)
             entry.model = record.model;
         if (record.ts) {
@@ -1300,11 +1306,19 @@ function syncUsageFromSessions(config, configPath, usagePath) {
             catch {
                 return;
             }
-            const resolved = resolveProfileForSession(config, logEntries, type, filePath, stats.sessionId);
-            if (!resolved.match)
-                return;
             const sessionKey = stats.sessionId ? buildSessionKey(type, stats.sessionId) : null;
             const sessionPrev = sessionKey ? sessions[sessionKey] : null;
+            const sessionProfile = sessionPrev && (sessionPrev.profileKey || sessionPrev.profileName)
+                ? {
+                    profileKey: sessionPrev.profileKey || null,
+                    profileName: sessionPrev.profileName || null,
+                }
+                : null;
+            const resolved = sessionProfile
+                ? { match: sessionProfile, ambiguous: false }
+                : resolveProfileForSession(config, logEntries, type, filePath, stats.sessionId);
+            if (!resolved.match)
+                return;
             const resolvedModel = (sessionPrev && sessionPrev.model) ||
                 stats.model ||
                 (prev && prev.model) ||
@@ -1397,6 +1411,8 @@ function syncUsageFromSessions(config, configPath, usagePath) {
                     endTs: stats.endTs || (sessionPrev ? sessionPrev.endTs : null),
                     cwd: stats.cwd || (sessionPrev ? sessionPrev.cwd : null),
                     model: resolvedModel,
+                    profileKey: resolved.match.profileKey,
+                    profileName: resolved.match.profileName,
                 };
             }
             files[filePath] = {

package/docs/README.md

@@ -0,0 +1,3 @@
+# Docs Index
+
+- [usage.md](usage.md): Usage tracking and statusline input details (Codex/Claude).

package/docs/README_zh.md

@@ -0,0 +1,3 @@
+# 文档索引
+
+- [usage_zh.md](usage_zh.md)：用量统计逻辑与状态栏输入说明。

package/docs/usage.md

@@ -0,0 +1,126 @@
+# Usage tracking
+
+Usage stats are derived from two sources (statusline input + session logs) and appended to `usage.jsonl`.
+
+## Files and paths
+
+- `usage.jsonl`: JSONL records with `ts`, `type`, `profileKey`/`profileName`, `model`, `sessionId`, and token breakdowns.
+- `usage.jsonl.state.json`: per-session totals + per-session-file metadata (mtime/size) used to compute deltas and avoid double counting.
+- `profile-log.jsonl`: profile usage + session binding log (`use` and `session` events).
+- `statusline-debug.jsonl`: optional debug capture when `CODE_ENV_STATUSLINE_DEBUG` is enabled.
+- Paths can be overridden via `usagePath`, `usageStatePath`, `profileLogPath`, `codexSessionsPath`, `claudeSessionsPath`.
+
+## Session binding (profile -> session)
+
+- `codenv init` installs a shell wrapper so `codex`/`claude` run via `codenv launch`.
+- `codenv launch` logs profile usage and then finds the latest unbound session file (prefers matching `cwd`, within a small grace window) and records the binding in `profile-log.jsonl`.
+- Sync uses these bindings to attribute session usage to a profile.
+
+## Statusline sync (`codenv statusline --sync-usage`)
+
+- Requires `sessionId`, model, and a profile (`profileKey` or `profileName`) to write usage.
+- Reads stdin JSON for totals and computes a delta against the last stored totals in the state file.
+- If totals decrease (session reset), the current totals are treated as fresh usage; otherwise negative sub-deltas are clamped to zero.
+- If the token breakdown exceeds the reported total, the breakdown sum wins.
+- Appends a delta record to `usage.jsonl` and updates the per-session totals in the state file.
+
+## Session log sync (`--sync-usage` and `codenv list`)
+
+- Scans Codex sessions under `CODEX_HOME/sessions` (or `~/.codex/sessions`) and Claude sessions under `CLAUDE_HOME/projects` (or `~/.claude/projects`).
+- Codex: parses `event_msg` token_count records and uses max totals; cached input is recorded as cache read and subtracted from input when possible.
+- Claude: sums `message.usage` input/output/cache tokens across the file.
+- Deltas are computed against prior file metadata and per-session maxima in the state file; files without a resolved binding are skipped.
+
+## Daily totals
+
+- "Today" is computed in local time between 00:00 and the next 00:00.
+
+## Cost calculation
+
+- Uses pricing from the profile (or `pricing.models`, plus defaults) and requires token splits; if splits are missing, cost is omitted.
+
+## Examples
+
+### `usage.jsonl` record
+
+Each line is a JSON object (fields may be null/omitted depending on the source):
+
+```json
+{"ts":"2026-01-25T12:34:56.789Z","type":"codex","profileKey":"p_a1b2c3","profileName":"primary","model":"gpt-5.1-codex","sessionId":"a6f9c4d8-1234-5678-9abc-def012345678","inputTokens":1200,"outputTokens":300,"cacheReadTokens":200,"cacheWriteTokens":0,"totalTokens":1700}
+```
+
+`totalTokens` is the max of the reported total and the breakdown sum, so it can be >= input + output + cache.
+
+## Statusline input examples (Codex/Claude)
+
+To see real payloads, set `CODE_ENV_STATUSLINE_DEBUG=1` and read the JSONL entries in
+`statusline-debug.jsonl` (or the path from `CODE_ENV_STATUSLINE_DEBUG_PATH`).
+
+### Codex (token_usage totals)
+
+```json
+{
+  "type": "codex",
+  "session_id": "a6f9c4d8-1234-5678-9abc-def012345678",
+  "profile": { "key": "p_a1b2c3", "name": "primary", "type": "codex" },
+  "model": "gpt-5.1-codex",
+  "token_usage": {
+    "total_token_usage": {
+      "input_tokens": 1200,
+      "output_tokens": 300,
+      "cached_input_tokens": 200,
+      "cache_creation_input_tokens": 50,
+      "total_tokens": 1750
+    },
+    "last_token_usage": {
+      "input_tokens": 100,
+      "output_tokens": 20,
+      "cached_input_tokens": 10,
+      "cache_creation_input_tokens": 0,
+      "total_tokens": 130
+    }
+  }
+}
+```
+
+Accepted aliases (non-exhaustive):
+- `token_usage.total_token_usage` or `token_usage.totalTokenUsage`
+- `last_token_usage` or `lastTokenUsage`
+- `input_tokens` / `inputTokens` / `input`
+- `output_tokens` / `outputTokens` / `output` / `reasoning_output_tokens`
+- `cached_input_tokens` / `cache_read_input_tokens`
+- `cache_creation_input_tokens` / `cache_write_input_tokens`
+- `total_tokens` / `totalTokens` / `total`
+
+`token_usage` can also be a number, or the payload can provide `usage` directly.
+
+### Claude (context_window totals)
+
+```json
+{
+  "type": "claude",
+  "session_id": "1f2e3d4c-5678-90ab-cdef-1234567890ab",
+  "profile": { "key": "p_c3d4e5", "name": "default", "type": "claude" },
+  "model": { "display_name": "Claude Sonnet 4.5" },
+  "context_window": {
+    "total_input_tokens": 800,
+    "total_output_tokens": 250,
+    "current_usage": {
+      "cache_read_input_tokens": 100,
+      "cache_creation_input_tokens": 40
+    },
+    "context_window_size": 200000
+  }
+}
+```
+
+Accepted aliases (non-exhaustive):
+- `context_window` or `contextWindow`
+- `current_usage` or `currentUsage`
+- `total_input_tokens` / `totalInputTokens`
+- `total_output_tokens` / `totalOutputTokens`
+- `cache_read_input_tokens` / `cacheReadInputTokens`
+- `cache_creation_input_tokens` / `cacheWriteInputTokens`
+
+`usage` can also be provided directly with `todayTokens` / `totalTokens` / `inputTokens` /
+`outputTokens` / `cacheReadTokens` / `cacheWriteTokens`.

package/docs/usage_zh.md

@@ -0,0 +1,126 @@
+# 用量统计逻辑
+
+用量来自两条路径（状态栏输入同步 + 会话日志解析），最终追加到 `usage.jsonl`。
+
+## 文件与路径
+
+- `usage.jsonl`：JSONL 记录，包含 `ts`/`type`/`profileKey`/`profileName`/`model`/`sessionId` 和 token 拆分字段。
+- `usage.jsonl.state.json`：保存每个 session 的累计 totals + session 文件的 mtime/size，用于计算增量并避免重复统计。
+- `profile-log.jsonl`：profile 使用与 session 绑定日志（`use`/`session`）。
+- `statusline-debug.jsonl`：当 `CODE_ENV_STATUSLINE_DEBUG` 开启时写入的调试信息。
+- 可通过 `usagePath`/`usageStatePath`/`profileLogPath`/`codexSessionsPath`/`claudeSessionsPath` 覆盖默认路径。
+
+## 会话绑定（profile -> session）
+
+- `codenv init` 安装 shell 包装函数，使 `codex`/`claude` 实际走 `codenv launch`。
+- `codenv launch` 记录 profile 使用，并在启动后短时间内（默认约 5 秒、每秒轮询）找到最新未绑定的 session 文件（优先 `cwd` 匹配），写入 `profile-log.jsonl`。
+- 后续同步会用该绑定将 session 归因到对应 profile。
+
+## 状态栏同步（`codenv statusline --sync-usage`）
+
+- 需要 `sessionId`、model，以及 profile（`profileKey` 或 `profileName`）才会写入用量。
+- 从 stdin JSON 读取 totals，与 state 中的上次 totals 做差得到增量。
+- 如果 totals 回退（session reset），直接把当前 totals 当作新增；否则对负数子项做 0 处理。
+- 当拆分合计大于 total 时，以拆分合计为准。
+- 把增量写入 `usage.jsonl`，并更新 state 中的 session totals。
+
+## 会话日志同步（`--sync-usage` 与 `codenv list`）
+
+- 扫描 Codex 的 `CODEX_HOME/sessions`（或 `~/.codex/sessions`）与 Claude 的 `CLAUDE_HOME/projects`（或 `~/.claude/projects`）。
+- Codex：读取 `event_msg` 的 token_count 记录，取累计最大值；cached input 计为 cache read，并在可判断时从 input 中扣除。
+- Claude：汇总 `message.usage` 中的 input/output/cache tokens。
+- 与 state 中的文件元数据和 session 最大值做差，生成增量并写入 `usage.jsonl`；无法解析到绑定的文件会被跳过。
+
+## 今日统计
+
+- “今日”按本地时区 00:00 到次日 00:00 计算。
+
+## 费用换算
+
+- 使用 profile 定价或 `pricing.models`（含默认值）换算，依赖 input/output/cache 拆分；缺少拆分则不显示金额。
+
+## 示例
+
+### `usage.jsonl` 记录
+
+每行一条 JSON（字段可能为空或缺省，取决于来源）：
+
+```json
+{"ts":"2026-01-25T12:34:56.789Z","type":"codex","profileKey":"p_a1b2c3","profileName":"primary","model":"gpt-5.1-codex","sessionId":"a6f9c4d8-1234-5678-9abc-def012345678","inputTokens":1200,"outputTokens":300,"cacheReadTokens":200,"cacheWriteTokens":0,"totalTokens":1700}
+```
+
+`totalTokens` 会取“上报 total”和“拆分合计”的较大值，因此可能 >= input + output + cache。
+
+## 状态栏输入示例（Codex/Claude）
+
+可通过设置 `CODE_ENV_STATUSLINE_DEBUG=1`，在 `statusline-debug.jsonl`
+（或 `CODE_ENV_STATUSLINE_DEBUG_PATH` 指定路径）看到实际 JSON。
+
+### Codex（token_usage totals）
+
+```json
+{
+  "type": "codex",
+  "session_id": "a6f9c4d8-1234-5678-9abc-def012345678",
+  "profile": { "key": "p_a1b2c3", "name": "primary", "type": "codex" },
+  "model": "gpt-5.1-codex",
+  "token_usage": {
+    "total_token_usage": {
+      "input_tokens": 1200,
+      "output_tokens": 300,
+      "cached_input_tokens": 200,
+      "cache_creation_input_tokens": 50,
+      "total_tokens": 1750
+    },
+    "last_token_usage": {
+      "input_tokens": 100,
+      "output_tokens": 20,
+      "cached_input_tokens": 10,
+      "cache_creation_input_tokens": 0,
+      "total_tokens": 130
+    }
+  }
+}
+```
+
+可识别字段（节选）：
+- `token_usage.total_token_usage` 或 `token_usage.totalTokenUsage`
+- `last_token_usage` 或 `lastTokenUsage`
+- `input_tokens` / `inputTokens` / `input`
+- `output_tokens` / `outputTokens` / `output` / `reasoning_output_tokens`
+- `cached_input_tokens` / `cache_read_input_tokens`
+- `cache_creation_input_tokens` / `cache_write_input_tokens`
+- `total_tokens` / `totalTokens` / `total`
+
+`token_usage` 也可以是数字，或直接提供 `usage`。
+
+### Claude（context_window totals）
+
+```json
+{
+  "type": "claude",
+  "session_id": "1f2e3d4c-5678-90ab-cdef-1234567890ab",
+  "profile": { "key": "p_c3d4e5", "name": "default", "type": "claude" },
+  "model": { "display_name": "Claude Sonnet 4.5" },
+  "context_window": {
+    "total_input_tokens": 800,
+    "total_output_tokens": 250,
+    "current_usage": {
+      "cache_read_input_tokens": 100,
+      "cache_creation_input_tokens": 40
+    },
+    "context_window_size": 200000
+  }
+}
+```
+
+可识别字段（节选）：
+- `context_window` 或 `contextWindow`
+- `current_usage` 或 `currentUsage`
+- `total_input_tokens` / `totalInputTokens`
+- `total_output_tokens` / `totalOutputTokens`
+- `cache_read_input_tokens` / `cacheReadInputTokens`
+- `cache_creation_input_tokens` / `cacheWriteInputTokens`
+
+也可直接提供 `usage`，字段包括 `todayTokens` / `totalTokens` / `inputTokens` /
+`outputTokens` / `cacheReadTokens` / `cacheWriteTokens`。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@praeviso/code-env-switch",
-  "version": "0.1.6",
+  "version": "0.1.7",
   "description": "Switch between Claude Code and Codex environment variables from a single CLI",
   "bin": {
     "codenv": "bin/index.js"

package/src/usage/index.ts

@@ -91,6 +91,8 @@ interface UsageSessionEntry {
     endTs: string | null;
     cwd: string | null;
     model?: string | null;
+    profileKey?: string | null;
+    profileName?: string | null;
 }
 
 interface UsageStateFile {
@@ -638,6 +640,8 @@ export function syncUsageFromStatuslineInput(
             endTs: now,
             cwd: cwd || (prev ? prev.cwd : null),
             model: resolvedModel,
+            profileKey: profileKey || null,
+            profileName: profileName || null,
         };
         state.sessions = sessions;
         updateUsageStateMetadata(state, usagePath);
@@ -1387,6 +1391,8 @@ function buildUsageSessionsFromRecords(
         entry.cacheReadTokens += toUsageNumber(record.cacheReadTokens);
         entry.cacheWriteTokens += toUsageNumber(record.cacheWriteTokens);
         entry.totalTokens += toUsageNumber(record.totalTokens);
+        if (record.profileKey) entry.profileKey = record.profileKey;
+        if (record.profileName) entry.profileName = record.profileName;
         if (!entry.model && record.model) entry.model = record.model;
         if (record.ts) {
             const range = { start: entry.startTs, end: entry.endTs };
@@ -1563,17 +1569,26 @@ export function syncUsageFromSessions(
             } catch {
                 return;
             }
-            const resolved = resolveProfileForSession(
-                config,
-                logEntries,
-                type,
-                filePath,
-                stats.sessionId
-            );
-            if (!resolved.match) return;
             const sessionKey =
                 stats.sessionId ? buildSessionKey(type, stats.sessionId) : null;
             const sessionPrev = sessionKey ? sessions[sessionKey] : null;
+            const sessionProfile =
+                sessionPrev && (sessionPrev.profileKey || sessionPrev.profileName)
+                    ? {
+                          profileKey: sessionPrev.profileKey || null,
+                          profileName: sessionPrev.profileName || null,
+                      }
+                    : null;
+            const resolved = sessionProfile
+                ? { match: sessionProfile, ambiguous: false }
+                : resolveProfileForSession(
+                      config,
+                      logEntries,
+                      type,
+                      filePath,
+                      stats.sessionId
+                  );
+            if (!resolved.match) return;
             const resolvedModel =
                 (sessionPrev && sessionPrev.model) ||
                 stats.model ||
@@ -1683,6 +1698,8 @@ export function syncUsageFromSessions(
                     endTs: stats.endTs || (sessionPrev ? sessionPrev.endTs : null),
                     cwd: stats.cwd || (sessionPrev ? sessionPrev.cwd : null),
                     model: resolvedModel,
+                    profileKey: resolved.match.profileKey,
+                    profileName: resolved.match.profileName,
                 };
             }
             files[filePath] = {