@cryer/star-cli 0.3.9 → 0.3.11

package/README.md

@@ -66,7 +66,7 @@ maxSteps = 100
 contextCompaction = "summary"     # summary | truncate — how over-budget history is compacted
 streamMaxRetries = 5              # extra attempts per model request on transient failures
                                   # (network error, 429/5xx, idle cutoff, empty reply); 0 disables
-streamIdleTimeoutSec = 20         # seconds with no stream output before a stalled response is
+streamIdleTimeoutSec = 60         # seconds with no stream output before a stalled response is
                                   # ended gracefully (some relays never close the stream)
 streamFirstChunkTimeoutSec = 300  # seconds to wait for the very first content part — slow
                                   # thinking-model endpoints can stay silent for minutes
@@ -177,7 +177,7 @@ model = "kimi-for-coding"
 temperature = 1        # mandatory here — the endpoint rejects any other value
 contextMaxTokens = 128000
 # raise these when the relay buffers long generations:
-# streamIdleTimeoutSec = 60        # default 20
+# streamIdleTimeoutSec = 120       # default 60
 # streamFirstChunkTimeoutSec = 600 # default 300
 ```
 
@@ -389,3 +389,12 @@ pnpm test:pipeline   # layered pipeline incl. optional live LLM smoke (needs an
 CI runs lint, typecheck, tests, and build on Ubuntu + Windows against Node 20/22 (`.github/workflows/ci.yml`).
 
 Architecture: `src/cli` (Ink UI), `src/agent` (main loop), `src/llm` (Vercel AI SDK provider layer), `src/tools`, `src/context` (token budget + compaction), `src/permissions`, `src/session`, `src/config`.
+
+## Benchmark
+
+[Terminal-Bench 2.0](https://github.com/laude-institute/terminal-bench), 20 randomly sampled tasks, same model for both agents:
+
+| Agent | Model | Pass rate | Avg tokens | Avg duration |
+|---|---|---|---|---|
+| Star CLI 0.3.9 | Kimi K3 | 17/20 (85%) | 129.2K | 414s |
+| terminus-2 2.0.0 | Kimi K3 | 9/20 (45%) | 49.5K | 484s |

package/README.zh-CN.md

@@ -65,7 +65,7 @@ maxSteps = 100
 contextCompaction = "summary"     # summary | truncate —— 历史超出预算时的压缩方式
 streamMaxRetries = 5              # 单次模型请求在瞬时失败（网络错误、429/5xx、idle 截断、
                                   # 空回复）时的额外重试次数；0 表示禁用
-streamIdleTimeoutSec = 20         # 流没有任何输出的秒数，超过后将停滞的响应优雅结束
+streamIdleTimeoutSec = 60         # 流没有任何输出的秒数，超过后将停滞的响应优雅结束
                                   #（有些中转站从不关闭流）
 streamFirstChunkTimeoutSec = 300  # 等待第一个内容分片的秒数——慢速思考模型端点在接受请求后
                                   # 可能静默数分钟
@@ -172,7 +172,7 @@ model = "kimi-for-coding"
 temperature = 1        # 此处必须设置——该端点拒绝任何其他取值
 contextMaxTokens = 128000
 # 中转站长时间缓冲整段生成时可调大：
-# streamIdleTimeoutSec = 60        # 默认 20
+# streamIdleTimeoutSec = 120       # 默认 60
 # streamFirstChunkTimeoutSec = 600 # 默认 300
 ```
 
@@ -384,3 +384,12 @@ pnpm test:pipeline   # layered pipeline incl. optional live LLM smoke (needs an
 CI 在 Ubuntu + Windows 上、Node 20/22 下运行 lint、typecheck、测试和构建（`.github/workflows/ci.yml`）。
 
 架构：`src/cli`（Ink UI）、`src/agent`（主循环）、`src/llm`（Vercel AI SDK 服务商层）、`src/tools`、`src/context`（token 预算 + 压缩）、`src/permissions`、`src/session`、`src/config`。
+
+## 基准测试
+
+[Terminal-Bench 2.0](https://github.com/laude-institute/terminal-bench)，随机抽取 20 个任务，两个 agent 使用相同模型：
+
+| Agent | 模型 | 通过率 | 平均 token 消耗 | 平均耗时 |
+|---|---|---|---|---|
+| Star CLI 0.3.9 | Kimi K3 | 17/20 (85%) | 129.2K | 414s |
+| terminus-2 2.0.0 | Kimi K3 | 9/20 (45%) | 49.5K | 484s |