opencode-tui-stats 0.4.1 → 0.4.3

package/README.en.md

@@ -1,6 +1,6 @@
 # opencode-tui-stats
 
-A TUI plugin for OpenCode 1.x (v1) that shows a single line — `TPS | AVG | TTFT` — on the right side of the session prompt.
+A TUI plugin for OpenCode 1.x (v1), mounted at the bottom of the right sidebar (slot `sidebar_footer`), with the panel title `Model-Performance · observed`. Each model occupies one row that can be collapsed/expanded by mouse click.
 
 ## Demo
 
@@ -18,11 +18,25 @@ This command writes the plugin into opencode's TUI config. You can also install
 
 - opencode ≥ 1.3.14
 
+## Collapsing & subagents
+
+- Each model occupies one row that can be collapsed/expanded by mouse click (`▼`/`▶`); rows are collapsed by default, and the most recently used model starts expanded.
+- Collapsed: `▶ [↳ ]provider/model` (redundant same-provider prefix omitted; provider dimmed, model name bright).
+- Expanded:
+  - `AVG …t/s · TPS …t/s` (TPS shown only when a value exists; the final value is kept and dimmed after generation ends)
+  - `TTFT …s · CACHE …%`
+  - `$…` (cost on its own line)
+- Models from subagents (sub-sessions) are merged into the same panel; rows originating from a subagent are prefixed with `↳`, and the same model (parent/child) is merged and summed.
+- Panel footer: `Session CACHE …%` (aggregated across the whole session + subagent tree, 1 decimal place); `retries` is shown only when >0.
+
 ## Metrics
 
-- `TPS`: live generation rate within the most recent ~5-second streaming window (tokens/sec); shows `-` when the session is idle or data is stale.
+- `TPS`: generation rate within the most recent ~5-second streaming window (tokens/sec); the final value is frozen and dimmed after the stream stops.
 - `AVG`: average generation rate of completed messages in this session (tokens/sec).
-- `TTFT`: average time to first token of completed messages in this session (seconds).
+- `TTFT`: average time to first token (seconds, approximate, includes queue/retry wait).
+- `CACHE`: prompt cache hit rate = `cache.read / (input + cache.read + cache.write)`, 1 decimal place.
+- `COST`: cumulative session cost (aggregated per model/session).
+- `RETRY`: retry count, derived from the attempt reported by the provider.
 
 ## License
 

package/README.md

@@ -1,6 +1,6 @@
 # opencode-tui-stats
 
-OpenCode 1.x（v1）的 TUI 插件，在会话提示区右侧显示一行 `TPS | AVG | TTFT`。
+OpenCode 1.x（v1）的 TUI 插件，安装在右侧栏底部（slot `sidebar_footer`），面板标题为 `Model-Performance · observed`。每个模型占一行，鼠标点击可折叠/展开。
 
 ## 演示
 
@@ -18,11 +18,25 @@ opencode plugin opencode-tui-stats --global
 
 - opencode ≥ 1.3.14
 
+## 折叠与子会话
+
+- 每个模型占一行，鼠标点击可折叠/展开（`▼`/`▶`）；默认折叠，最近使用的模型默认展开。
+- 折叠态：`▶ [↳ ]provider/model`（同 provider 前缀省略；provider 暗色、模型名亮色）。
+- 展开态：
+  - `AVG …t/s · TPS …t/s`（TPS 仅在有值时显示；生成结束后保留最后值并变暗）
+  - `TTFT …s · CACHE …%`
+  - `$…`（成本单独一行）
+- 子会话（子 agent）的模型会并入同一面板，来自子会话的行前缀 `↳`；同一模型（父/子）合并求和。
+- 面板底部：`Session CACHE …%`（整个会话 + 子会话树聚合，1 位小数）；`retries` 仅在 >0 时显示。
+
 ## 指标说明
 
-- `TPS`：最近约 5 秒流式窗口内的实时生成速度（token/秒）；会话空闲或数据过期时显示 `-`。
+- `TPS`：最近约 5 秒流式窗口内的生成速度（token/秒）；停止后冻结最后值并变暗。
 - `AVG`：本会话已完成消息的平均生成速度（token/秒）。
-- `TTFT`：本会话已完成消息的平均首字延迟（秒）。
+- `TTFT`：平均首字延迟（秒，近似值，包含排队/重试等待）。
+- `CACHE`：提示词缓存命中率 = `cache.read / (input + cache.read + cache.write)`，1 位小数。
+- `COST`：会话累计成本（按模型/会话聚合）。
+- `RETRY`：按 provider 上报的 attempt 统计的重试次数。
 
 ## License