fix(openclaw-plugin): capture user turn at assistant boundary

[yeyitech]

Summary

• backtrack the effective afterTurn capture window when prePromptMessageCount starts on the assistant side of the turn
• keep user-only <relevant-memories> stripping while preserving assistant text
• add regression coverage for assistant-boundary and grouped-user cases

Closes #1248

Testing

• pnpm vitest run tests/ut/context-engine-afterTurn.test.ts

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis ✅ 1248 - PR Code Verified Compliant requirements: Fix afterTurn auto-capture to include user message Capture both user and assistant messages from current turn Requires further human verification: Ensure VLM extractor receives full turn (user + assistant)

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🏅 Score: 80

🧪 PR contains tests

🔒 No security concerns identified

✅ No TODO sections

🔀 No multiple PR themes

⚡ Recommended focus areas for review Unused Counter Variable The count variable in extractNewTurnMessages is declared but never incremented, so newCount is always 0. This affects diagnostics and runLocalPrecheck which rely on accurate new message count. let count = 0; const effectiveStartIndex = findEffectiveTurnStartIndex(messages, startIndex); // First pass: collect toolUse inputs indexed by toolCallId/toolUseId // Scan all messages (including after startIndex) to find toolUse before each toolResult const toolUseInputs: Record<string, Record<string, unknown>> = {}; for (let i = 0; i < messages.length; i++) { const msg = messages[i] as Record<string, unknown>; if (!msg || typeof msg !== "object") continue; const role = msg.role as string; if (role === "assistant") { const content = msg.content; if (Array.isArray(content)) { for (const block of content) { const b = block as Record<string, unknown>; // Handle toolCall, toolUse, tool_call types if (b?.type === "toolCall" || b?.type === "toolUse" || b?.type === "tool_call") { const id = (b.id as string) || (b.toolUseId as string) || (b.toolCallId as string); // Try multiple field names for tool input: arguments, input, toolInput const input = b.arguments ?? b.input ?? b.toolInput; if (id && input && typeof input === "object") { toolUseInputs[id] = input as Record<string, unknown>; } } } } } } for (let i = effectiveStartIndex; i < messages.length; i++) { const msg = messages[i] as Record<string, unknown>; if (!msg || typeof msg !== "object") continue; const role = msg.role as string; if (!role || role === "system") continue; count++; // toolResult -> type: "tool" if (role === "toolResult") { const toolName = typeof msg.toolName === "string" ? msg.toolName : "tool"; const output = formatToolResultContent(msg.content) || ""; // Try multiple field names for tool call ID const toolCallId = (msg.toolCallId as string) || (msg.toolUseId as string) || (msg.tool_call_id as string); const toolInput = toolCallId && toolUseInputs[toolCallId] ? toolUseInputs[toolCallId] : (typeof msg.toolInput === "object" && msg.toolInput !== null ? msg.toolInput as Record<string, unknown> : undefined); if (output) { pushExtractedPart(result, "user", { type: "tool", toolCallId: toolCallId || undefined, toolName, toolInput, toolOutput: `[${toolName} result]: ${output}`, toolStatus: "completed", }); } continue; } // user/assistant -> type: "text" const text = extractSingleMessageText(msg); if (text) { const cleanedText = role === "assistant" ? (HEARTBEAT_RE.test(text) ? "" : text.trim()) : sanitizeUserTextForCapture(text); if (cleanedText) { const ovRole: "user" | "assistant" = role === "assistant" ? "assistant" : "user"; pushExtractedPart(result, ovRole, { type: "text", text: cleanedText, }); } } } return { messages: result, newCount: count, effectiveStartIndex }; Unintended Tool Output Change Tool result output is now wrapped in [${toolName} result]: ${output} without mention in PR summary. This may break existing extraction logic that expects raw tool output. pushExtractedPart(result, "user", { type: "tool", toolCallId: toolCallId || undefined, toolName, toolInput, toolOutput: `[${toolName} result]: ${output}`, toolStatus: "completed", });

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Stop backward scan at previous turn Break the backward loop when encountering a non-user, non-system role (e.g., "assistant", "toolResult") to prevent capturing user messages from earlier turns. This ensures we only look back within the current turn boundary. examples/openclaw-plugin/text-utils.ts [524-545] for (let i = firstRelevantIndex - 1; i >= 0; i -= 1) { const msg = messages[i] as Record<string, unknown>; const role = typeof msg?.role === "string" ? msg.role : ""; if (!role || role === "system") { continue; } if (role === "user") { let earliestUserIndex = i; for (let j = i - 1; j >= 0; j -= 1) { const prev = messages[j] as Record<string, unknown>; const prevRole = typeof prev?.role === "string" ? prev.role : ""; if (!prevRole || prevRole === "system") { continue; } if (prevRole !== "user") { break; } earliestUserIndex = j; } return earliestUserIndex; } + // Stop at the first non-user, non-system message (previous turn boundary) + break; } Suggestion importance[1-10]: 7 __ Why: Adding a break when encountering a non-user, non-system role prevents capturing user messages from earlier turns, ensuring correct turn boundary detection.	Medium