[Bug] "Prompt is too long" at ~254K tokens despite Claude Code supporting 1M context

[wxdmczy]

Environment

• OS: Windows 11 Pro
• cc-connect: v1.2.2-beta.4 (npm beta)
• Agent: claudecode
• Platform: feishu (WebSocket)
• Claude Code: latest, with 1M context window enabled

Problem

When using cc-connect with Claude Code via Feishu, the session hits "prompt is too long" error at approximately 254K tokens. However, my Claude Code CLI supports 1M context window natively.

Expected Behavior

cc-connect should utilize the full 1M context window available to Claude Code, or at least handle context limits more gracefully (e.g., auto-compress before hitting the limit).

Actual Behavior

The session fails with "prompt is too long" well before reaching the 1M context limit. The error forces me to manually run /new to start a fresh session.

Analysis

Looking at the source code (agent/claudecode/session.go), cc-connect spawns Claude Code with:

args := []string{
    "--output-format", "stream-json",
    "--input-format", "stream-json",
    "--permission-prompt-tool", "stdio",
}

No context window or token limit flags are passed. PR #137 added fallback behavior (auto-recover to fresh session on "prompt is too long"), but the root cause — why the limit is hit at ~254K instead of 1M — is not addressed.

Questions

1. Is there an internal limit in cc-connect that caps context below Claude Code's native 1M?
2. Should cc-connect pass --max-context-tokens or similar flag to Claude Code CLI?
3. Could the [auto_compress] config section help here? If so, what's the recommended configuration?

Config

[[projects]]
name = "feishu-workspace"

[projects.agent]
type = "claudecode"

[projects.agent.options]
work_dir = "C:\\Users\\zy"
mode = "default"

[[projects.platforms]]
type = "feishu"

[projects.heartbeat]
[projects.auto_compress]

Workaround

Currently using /new to start fresh sessions when context fills up.

[github-actions]

👋 Hi @wxdmczy! Thanks for opening your first issue here!

We'll review it as soon as possible. While waiting, please make sure:

• You've provided enough details about the issue
• Any error messages are included in full
• For feature requests, describe your use case

Thanks for your feedback!

[chenhg5]

感谢详细反馈！

分析结论：
cc-connect 本身不限制 context，问题可能在 Claude Code CLI 配置。

建议方案：

1. 检查 Claude Code CLI 是否需要 --max-context-tokens 参数来启用 1M context
2. 在 config.toml 中启用 [projects.auto_compress] 配置，可以在接近限制时自动压缩

后续改进：
我们会考虑在 claudecode agent options 中添加 context_tokens 配置项，允许用户指定 context 大小。

如果找到解决方案，欢迎分享！