fix: remove 30-step limit — match original usize::MAX (conversation.rs)

Original claw-code conversation.rs line 141:
max_iterations: usize::MAX

This means the agent loop runs until the LLM stops calling tools,
with NO artificial step limit. The 30-step limit was a Claw-Web
deviation that broke long-running tasks.

Also fixed comments about compaction — it IS manual-only in both
original and Claw-Web (via /compact command).

server/runtime/agent.ts

@@ -6,14 +6,14 @@
 import { ENV } from "../_core/env";
 import { buildSystemPrompt, TOOL_DEFINITIONS } from "./system-prompt";
 import { executeTool, getPlanMode, runPreToolHooks, runPostToolHooks, initializeMcpFromConfig, getMcpManager } from "../tools/executor";
-// Compact is manual-only via /compact command (matches original claw-code)
-// import { shouldCompact, compactSession, dbMessagesToSession, DEFAULT_COMPACTION_CONFIG } from "./compact";
+// Compact is manual-only via /compact command (matches original claw-code conversation.rs)
 import { UsageTracker, pricingForModel, defaultSonnetTierPricing, estimateCostUsdWithPricing, totalCostUsd, formatUsd, summaryLinesForModel } from "./usage";
 import type { TokenUsage } from "./usage";
 import type { Response } from "express";
 import { execSync } from "child_process";
 
-// No auto-compaction — in original claw-code, compact is ONLY triggered by /compact command
+// In original claw-code, max_iterations defaults to usize::MAX (effectively unlimited).
+// Compact is triggered externally by the caller, not inside the loop.
 
 interface AgentMessage {
   role: "system" | "user" | "assistant" | "tool";
@@ -255,7 +255,9 @@ export async function runAgentLoop(
   let totalCompletionTokens = 0;
   let totalCost = 0;
   let iterations = 0;
-  const MAX_ITERATIONS = cfg.maxIterations || 30;
+  // Original claw-code: max_iterations defaults to usize::MAX (conversation.rs line 141)
+  // This means the loop runs until the LLM stops calling tools — no artificial step limit.
+  const MAX_ITERATIONS = cfg.maxIterations || Number.MAX_SAFE_INTEGER;
   const assistantMessages: AgentMessage[] = [];
   const toolResultMessages: AgentMessage[] = [];
 
@@ -516,7 +518,7 @@ export async function runAgentLoop(
         toolResultMessages.push(toolResultMsg);
       }
 
-      // No auto-compaction — in original claw-code, compact is ONLY manual via /compact command
+      // Compact is manual-only via /compact command (matches original claw-code conversation.rs)
 
       // Continue the loop — LLM will see tool results and decide next action
       sendSSE(res, "status", {
@@ -534,7 +536,7 @@ export async function runAgentLoop(
   }
 
   if (iterations >= MAX_ITERATIONS) {
-    sendSSE(res, "error", { message: `Maximum iterations (${MAX_ITERATIONS}) reached. The agent has been stopped to prevent runaway execution.` });
+    sendSSE(res, "error", { message: `Maximum iterations (${MAX_ITERATIONS}) reached. Use /compact to reduce context and continue.` });
   }
 
   // Calculate cost using UsageTracker (matches original)