feat: stream zeroclaw sidecar responses via ChatDelta events by Keith-CY · Pull Request #34 · lay2dev/clawpal

Keith-CY · 2026-03-02T07:31:16Z

Summary

Add async streaming execution path for zeroclaw sidecar responses — stdout is read line-by-line and emitted as ChatDelta events in real time, replacing the fully-buffered Command::output() approach
Add sanitize_line() per-line filter with shared OnceLock<Regex> patterns, refactoring sanitize_output() to share the same helpers
Add start_streaming() / send_streaming() inherent async methods on both ZeroclawDoctorAdapter and ZeroclawInstallAdapter (sync trait stays unchanged)
Wire streaming into all 5 command handlers: doctor_start_diagnosis, doctor_send_message, doctor_approve_invoke, install_start_session, install_send_message

No frontend changes needed — RuntimeEvent::ChatDelta already exists in the type system, doctor_runtime_bridge already maps it to doctor:chat-delta, and useDoctorAgent already handles it with replacement semantics.

Closes #26

Test plan

cargo build passes
cargo test — all 131 lib tests + 9 new sanitize_line integration tests pass (1 pre-existing unrelated failure excluded)
Manual: start a doctor diagnosis and observe progressive text appearing in the chat bubble
Manual: verify tool-call invocations still work (tool intent parsed from final text)
Manual: verify provider retry works (misconfigure first provider, confirm second streams correctly)

🤖 Generated with Claude Code

Zeroclaw sidecar responses were fully buffered — users saw no feedback during LLM generation. Add an async streaming execution path that reads stdout line-by-line and emits ChatDelta events as lines arrive, then emits ChatFinal when the process completes. - Add `sanitize_line()` for per-line filtering with shared OnceLock regex - Add `run_zeroclaw_message_streaming()` async fn using tokio::process - Add `start_streaming()` / `send_streaming()` on both adapters - Wire streaming into doctor and install command handlers - Add 9 new sanitize_line tests No frontend changes needed — the existing `useDoctorAgent` hook already handles `doctor:chat-delta` events with replacement semantics. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

dev01lay2

PR #34 Review — Streaming zeroclaw sidecar responses

Overall solid — line-by-line streaming via BufReader::lines() + ChatDelta is the right approach, and the sanitize refactor with shared OnceLock patterns is clean. A few issues worth addressing:

src-tauri/src/runtime/zeroclaw/process.rs

src-tauri/src/runtime/zeroclaw/adapter.rs

src-tauri/src/runtime/zeroclaw/process.rs

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 259238390c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

src-tauri/src/runtime/zeroclaw/process.rs

dev01lay2

Code looks great — streaming 架构清晰，retry 逻辑的抽取很干净，sanitize_line 的拆分和测试覆盖都到位。

唯一的问题：CI 的 cargo fmt --check 挂了，有两处格式问题需要修一下。跑个 cargo fmt --all 就好。

dev01lay2 · 2026-03-02T08:06:37Z

src-tauri/src/runtime/zeroclaw/mod.rs

@@ -1,4 +1,5 @@
 pub mod adapter;
+mod streaming;


BS: cargo fmt 要求 mod streaming 按字母序排在 pub mod session 之后。跑 cargo fmt --all 会自动修复。

dev01lay2 · 2026-03-02T08:06:37Z

src-tauri/src/runtime/zeroclaw/streaming.rs

+        append_history(&session_key, "user", message);
+    }
+
+    let text = run_zeroclaw_message_streaming(prompt, &key.instance_id, &key.storage_key(), on_delta)


BS: 这行超长，cargo fmt 会把它拆成多行。跑 cargo fmt --all 即可。

dev01lay2 · 2026-03-02T08:06:37Z

src-tauri/src/runtime/zeroclaw/process.rs

+    }
+
+    let status = child
+        .wait()


NBS: stream_once 每次 delta 发送完整累积文本而非增量 — 对于长回复可能有性能开销，但作为初版实现完全合理，后续可以优化为只发增量。

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1348570983

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-03-02T08:09:38Z

src-tauri/src/runtime/zeroclaw/install_adapter.rs

+        F: Fn(&str) + Send + 'static,
+    {
+        let session_key = key.storage_key();
+        append_history(&session_key, "user", message);


Avoid double-writing install user turns to history

send_streaming appends the user message before building the prompt, and then passes Some(message) into run_zeroclaw_streaming_turn, which appends the same user turn again (streaming.rs does this in its user_message branch). This means every install follow-up is stored twice, so subsequent prompts contain duplicated user history and quickly bloat/warp the install conversation state.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-03-02T08:09:38Z

src-tauri/src/runtime/zeroclaw/process.rs

+    let runtime =
+        tokio::runtime::Runtime::new().map_err(|e| format!("failed to initialize runtime: {e}"))?;
+    runtime.block_on(run_zeroclaw_retry(


Remove nested Tokio runtime from sync zeroclaw execution

run_zeroclaw_message now creates a new Tokio runtime and calls block_on, but this function is still called from async command handlers (for example explain_operation_error in agent_fallback.rs). In normal configured environments (where provider/env checks pass), this can panic with Tokio's "Cannot start a runtime from within a runtime" and turn recoverable fallback guidance requests into hard runtime failures.

Useful? React with 👍 / 👎.

Keith-CY · 2026-03-02T08:41:11Z

Added passphrase retry follow-up on top of #34: commit 3a7eaa9 fixes SSH passphrase fallback when host auth errors are retryable even if host entry not yet in local SSH list; also broadens retry hint matching for password/key-encrypted errors and adds unit coverage in .

Keith-CY · 2026-03-02T08:41:31Z

I added the follow-up test coverage and pushed commit 3a7eaa9 on this branch: src/App.tsx, src/lib/sshConnectErrors.ts, clawpal-core/src/ssh/mod.rs, src/lib/tests/sshConnectErrors.test.ts. If you want a separate PR, I can rebase/cherry-pick this commit onto a fresh branch, but this is already included in PR #34 now.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 57cc7f6602

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-03-02T08:57:04Z

src-tauri/src/runtime/zeroclaw/adapter.rs

+        .await?;
+        let session_key = key.storage_key();
+        append_history(&session_key, "system", &prompt);


Persist system prompt before streaming the first turn

In start_streaming, the call to run_zeroclaw_streaming_turn completes before the system prompt is written to history, but run_zeroclaw_streaming_turn itself appends the assistant reply immediately (streaming.rs), so the stored transcript order becomes assistant→system instead of system→assistant. Subsequent build_prompt_with_history* calls replay that reversed order, which can distort later turns because prior assistant content is shown before the governing system instruction; the same ordering bug appears in both doctor and install streaming start paths.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-03-02T08:57:04Z

src-tauri/src/runtime/zeroclaw/process.rs

+    let on_delta: std::sync::Arc<dyn Fn(&str) + Send + Sync> =
+        std::sync::Arc::new(on_delta);
+    (on_delta.as_ref())("");
+    run_zeroclaw_retry(


Clear streamed text between provider retry attempts

run_zeroclaw_message_streaming emits an empty delta only once before entering retry logic, so if an early provider/model attempt streams partial stdout and then fails, later retries run without resetting the visible buffer. When subsequent attempts fail before producing new stdout, the frontend can retain stale partial text from a failed attempt while the turn ends with an error, effectively presenting failed-attempt output as the visible assistant response.

Useful? React with 👍 / 👎.

dev01lay2

代码改动都很好 👍 — 之前提的 stdout/stderr 并发读取、retry 逻辑抽取、adapter 去重全都处理干净了。SSH passphrase retry 的改动也合理。

唯一的问题：CI 的 cargo fmt --check 还是挂了，process.rs 和 streaming.rs 里各有一处格式问题。跑个 cargo fmt --all 再 push 就行。

dev01lay2 · 2026-03-02T09:33:17Z

src-tauri/src/runtime/zeroclaw/process.rs

+            let args = args;
+            async move {
+                stream_once(&cmd, &cfg, &env_pairs, &args, on_delta.as_ref()).await
+            }


BS: 这个 closure 的缩进 cargo fmt 要求压缩成更紧凑的写法（CI 日志里有 expected diff）。cargo fmt --all 会自动修。

dev01lay2 · 2026-03-02T09:33:17Z

src-tauri/src/runtime/zeroclaw/streaming.rs

+        append_history(&session_key, "user", message);
+    }
+
+    let text = run_zeroclaw_message_streaming(prompt, &key.instance_id, &key.storage_key(), on_delta)


BS: 同上次 — 这行超长，cargo fmt 会拆成多行。cargo fmt --all 即可。

Finding 3: TokenBadge was using the global runtimeModel for cost estimation, ignoring per-session model overrides set via ModelSwitcher. Changes: - Doctor.tsx tracks sessionModelOverride state, derives effectiveModel (override ?? runtimeModel), passes it to TokenBadge - ModelSwitcher accepts onModelChange callback to notify parent when the user switches model or clears override - TokenBadge now receives the effective model so cost estimates match the actual model being used for the session Note: Findings 1 & 2 (streaming runtime override + usage) reference run_zeroclaw_message_streaming / stream_once which exist only in PR #34 (feat/zeroclaw-streaming-response), not in this branch's base. Will wire session override + usage into the streaming path when #34 merges.

dev01lay2 · 2026-03-03T01:50:32Z

📊 Test Coverage Report

Metric	Base (`main`)	PR (`feat/zeroclaw-streaming-response`)	Delta
Lines	62.76% (3912/6233)	63.05% (3923/6222)	🟢 +0.29%
Functions	57.69% (454/787)	57.99% (457/788)	🟢 +0.30%
Regions	62.89% (6324/10055)	63.07% (6338/10049)	🟢 +0.18%

Coverage measured by cargo llvm-cov (clawpal-core + clawpal-cli). Updated at 2026-03-03T01:50:30Z.

dev01lay2 · 2026-03-03T02:05:34Z

🧪 Test Coverage Report

PR #34: feat: stream zeroclaw sidecar responses via ChatDelta events
Branch: feat/zeroclaw-streaming-response → main

Rust Tests (clawpal-core)

	main (base)	feat/zeroclaw-streaming-response (PR)	Δ
✅ Pass	139	140	1
❌ Fail	1	0	-1
⏭️ Skip	0		0

TypeScript Tests (bun)

	main (base)	feat/zeroclaw-streaming-response (PR)	Δ
✅ Pass	9		-9
❌ Fail	0		0
📊 Line Coverage	100.00%	N/A%	—

Updated: 2026-03-03 02:36 UTC · clawpal-pr-coverage bot

dev01lay2 reviewed Mar 2, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Mar 2, 2026

View reviewed changes

src-tauri/src/runtime/zeroclaw/process.rs Show resolved Hide resolved

Keith-CY added 2 commits March 2, 2026 17:03

fix: align zeroclaw streaming with review feedback

5c5d846

fix: share zeroclaw retry loop between sync and streaming

1348570

dev01lay2 requested changes Mar 2, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Mar 2, 2026

View reviewed changes

fix: retry ssh passphrase flow when host not preloaded

3a7eaa9

Keith-CY force-pushed the feat/zeroclaw-streaming-response branch from d042eb2 to 799b79c Compare March 2, 2026 08:46

chore: align zeroclaw streaming closure traits and retry call signatures

57cc7f6

Keith-CY force-pushed the feat/zeroclaw-streaming-response branch from 799b79c to 57cc7f6 Compare March 2, 2026 08:50

chatgpt-codex-connector bot reviewed Mar 2, 2026

View reviewed changes

dev01lay2 requested changes Mar 2, 2026

View reviewed changes

This was referenced Mar 2, 2026

feat: in-session model switching and cost awareness (#32) #35

Merged

feat: structured diagnosis result cards #39

Merged

Conversation

Keith-CY commented Mar 2, 2026

Summary

Test plan

Uh oh!

dev01lay2 left a comment

Choose a reason for hiding this comment

PR #34 Review — Streaming zeroclaw sidecar responses

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

dev01lay2 left a comment

Choose a reason for hiding this comment

Uh oh!

dev01lay2 Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

dev01lay2 Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

dev01lay2 Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

Keith-CY commented Mar 2, 2026

Uh oh!

Keith-CY commented Mar 2, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

dev01lay2 left a comment

Choose a reason for hiding this comment

Uh oh!

dev01lay2 Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

dev01lay2 Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

dev01lay2 commented Mar 3, 2026

📊 Test Coverage Report

Uh oh!

dev01lay2 commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🧪 Test Coverage Report

Rust Tests (clawpal-core)

TypeScript Tests (bun)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

dev01lay2 commented Mar 3, 2026 •

edited

Loading