refactor(compaction): remove minMessages guard, use token-only threshold

Message count-based compaction trigger (default 200) was a legacy safety
net from when token estimation was unreliable. With calibrated overhead
estimation using actual LLM PromptTokens, the token threshold alone is
sufficient — matching industry standard (Claude Code, Anthropic API,
LangChain all use token-only).

Also fixes EstimateTokens vs EstimateHistoryTokens inconsistency in
end-of-run overhead calculation (loop_history.go), and updates stale
CLAUDE.md docs from 75% to 85%.

CLAUDE.md

@@ -57,7 +57,7 @@ ui/web/                       React SPA (pnpm, Vite, Tailwind, Radix UI)
 - **Agent types:** `open` (per-user context, 7 files) vs `predefined` (shared context + USER.md per-user)
 - **Context files:** `agent_context_files` (agent-level) + `user_context_files` (per-user), routed via `ContextFileInterceptor`
 - **Providers:** Anthropic (native HTTP+SSE), OpenAI-compat (HTTP+SSE), DashScope (Alibaba Qwen), Claude CLI (stdio+MCP bridge), ACP (Anthropic Console Proxy), Codex (OpenAI). All use `RetryDo()` for retries. Loads from `llm_providers` table with encrypted API keys
-- **Agent loop:** `RunRequest` → think→act→observe → `RunResult`. Events: `run.started`, `run.completed`, `chunk`, `tool.call`, `tool.result`. Auto-summarization at >75% context
+- **Agent loop:** `RunRequest` → think→act→observe → `RunResult`. Events: `run.started`, `run.completed`, `chunk`, `tool.call`, `tool.result`. Auto-summarization at >85% context (token-based only)
 - **Context propagation:** `store.WithAgentType(ctx)`, `store.WithUserID(ctx)`, `store.WithAgentID(ctx)`, `store.WithLocale(ctx)`
 - **WebSocket protocol (v3):** Frame types `req`/`res`/`event`. First request must be `connect`
 - **Config:** JSON5 at `GOCLAW_CONFIG` env. Secrets in `.env.local` or env vars, never in config.json

internal/agent/loop_history.go

@@ -553,18 +553,15 @@ func (l *Loop) maybeSummarize(ctx context.Context, sessionKey string) {
 	}
 	tokenEstimate := EstimateTokensWithCalibration(history, adjustedLastPT, lastMC)
 
-	// Resolve compaction thresholds from config with sensible defaults.
+	// Resolve compaction threshold from config: token-only (no message count guard).
+	// Industry standard — Claude Code, Anthropic API, LangChain all use token-based thresholds.
 	historyShare := config.DefaultHistoryShare
 	if l.compactionCfg != nil && l.compactionCfg.MaxHistoryShare > 0 {
 		historyShare = l.compactionCfg.MaxHistoryShare
 	}
-	minMessages := 200
-	if l.compactionCfg != nil && l.compactionCfg.MinMessages > 0 {
-		minMessages = l.compactionCfg.MinMessages
-	}
 
 	threshold := int(float64(l.contextWindow) * historyShare)
-	if len(history) <= minMessages && tokenEstimate <= threshold {
+	if tokenEstimate <= threshold {
 		return
 	}
 
@@ -681,7 +678,7 @@ func (l *Loop) estimateOverhead(history []providers.Message, lastPromptTokens, l
 	if count > len(history) {
 		count = len(history)
 	}
-	historyEstAtCalibration := EstimateTokens(history[:count])
+	historyEstAtCalibration := EstimateHistoryTokens(history[:count])
 	overhead := lastPromptTokens - historyEstAtCalibration
 	if overhead < 0 {
 		overhead = 0

internal/config/config.go

@@ -133,8 +133,7 @@ type AgentDefaults struct {
 // Matching TS agents.defaults.compaction.
 type CompactionConfig struct {
 	ReserveTokensFloor int                `json:"reserveTokensFloor,omitempty"` // min reserve tokens (default 20000)
-	MaxHistoryShare    float64            `json:"maxHistoryShare,omitempty"`    // max share of context for history (default 0.75)
-	MinMessages        int                `json:"minMessages,omitempty"`        // min messages before compaction triggers (default 200)
+	MaxHistoryShare    float64            `json:"maxHistoryShare,omitempty"`    // max share of context for history (default 0.85)
 	KeepLastMessages   int                `json:"keepLastMessages,omitempty"`   // messages to keep after compaction (default 4)
 	MemoryFlush        *MemoryFlushConfig `json:"memoryFlush,omitempty"`        // pre-compaction flush
 }

ui/desktop/frontend/src/i18n/locales/en/agents.json

@@ -344,8 +344,6 @@
       "description": "Context window compaction and memory flush settings",
       "maxHistoryShare": "Max History Share (0-1)",
       "maxHistoryShareTip": "Maximum fraction of context window for conversation history (e.g. 0.75 = 75%). Compaction triggers when exceeded.",
-      "minMessages": "Min Messages",
-      "minMessagesTip": "Minimum messages before compaction can trigger, regardless of token count.",
       "keepLastMessages": "Keep Last Messages",
       "keepLastMessagesTip": "Recent messages kept after compaction. Older messages are replaced by a summary.",
       "memoryFlush": "Memory Flush",

ui/desktop/frontend/src/i18n/locales/vi/agents.json

@@ -344,8 +344,6 @@
       "description": "Cài đặt nén cửa sổ ngữ cảnh và ghi nhớ trước nén",
       "maxHistoryShare": "Phần lịch sử tối đa (0-1)",
       "maxHistoryShareTip": "Tỷ lệ tối đa của cửa sổ ngữ cảnh dành cho lịch sử hội thoại (vd: 0.75 = 75%). Nén kích hoạt khi vượt quá.",
-      "minMessages": "Tin nhắn tối thiểu",
-      "minMessagesTip": "Số tin nhắn tối thiểu trước khi nén có thể kích hoạt, bất kể số token.",
       "keepLastMessages": "Giữ tin nhắn cuối",
       "keepLastMessagesTip": "Số tin nhắn gần nhất giữ lại sau khi nén. Các tin nhắn cũ hơn được thay thế bằng bản tóm tắt.",
       "memoryFlush": "Ghi nhớ trước nén",

ui/desktop/frontend/src/i18n/locales/zh/agents.json

@@ -344,8 +344,6 @@
       "description": "上下文窗口压缩和压缩前记忆设置",
       "maxHistoryShare": "最大历史份额 (0-1)",
       "maxHistoryShareTip": "上下文窗口用于对话历史的最大比例（如0.75 = 75%）。超过时触发压缩。",
-      "minMessages": "最小消息数",
-      "minMessagesTip": "压缩触发前所需的最少消息数，与令牌数无关。",
       "keepLastMessages": "保留最后消息数",
       "keepLastMessagesTip": "压缩后保留的最近消息数。旧消息被摘要替换。",
       "memoryFlush": "压缩前记忆",

ui/desktop/frontend/src/types/agent.ts

@@ -17,7 +17,6 @@ export interface MemoryConfig {
 export interface CompactionConfig {
   reserveTokensFloor?: number
   maxHistoryShare?: number
-  minMessages?: number
   keepLastMessages?: number
   memoryFlush?: {
     enabled?: boolean

ui/web/src/i18n/locales/en/agents.json

@@ -591,8 +591,6 @@
       "description": "Context window compaction and memory flush settings",
       "maxHistoryShare": "Max History Share (0-1)",
       "maxHistoryShareTip": "Maximum fraction of context window for conversation history (e.g. 0.75 = 75%). Compaction triggers when exceeded.",
-      "minMessages": "Min Messages",
-      "minMessagesTip": "Minimum messages before compaction can trigger, regardless of token count.",
       "keepLastMessages": "Keep Last Messages",
       "keepLastMessagesTip": "Recent messages kept after compaction. Older messages are replaced by a summary.",
       "memoryFlush": "Memory Flush",

ui/web/src/i18n/locales/vi/agents.json

@@ -591,8 +591,6 @@
       "description": "Cài đặt nén cửa sổ ngữ cảnh và ghi nhớ trước nén",
       "maxHistoryShare": "Phần lịch sử tối đa (0-1)",
       "maxHistoryShareTip": "Tỷ lệ tối đa của cửa sổ ngữ cảnh dành cho lịch sử hội thoại (vd: 0.75 = 75%). Nén kích hoạt khi vượt quá.",
-      "minMessages": "Tin nhắn tối thiểu",
-      "minMessagesTip": "Số tin nhắn tối thiểu trước khi nén có thể kích hoạt, bất kể số token.",
       "keepLastMessages": "Giữ tin nhắn cuối",
       "keepLastMessagesTip": "Số tin nhắn gần nhất giữ lại sau khi nén. Các tin nhắn cũ hơn được thay thế bằng bản tóm tắt.",
       "memoryFlush": "Ghi nhớ trước nén",

ui/web/src/i18n/locales/zh/agents.json

@@ -591,8 +591,6 @@
       "description": "上下文窗口压缩和压缩前记忆设置",
       "maxHistoryShare": "最大历史份额 (0-1)",
       "maxHistoryShareTip": "上下文窗口用于对话历史的最大比例（如0.75 = 75%）。超过时触发压缩。",
-      "minMessages": "最小消息数",
-      "minMessagesTip": "压缩触发前所需的最少消息数，与令牌数无关。",
       "keepLastMessages": "保留最后消息数",
       "keepLastMessagesTip": "压缩后保留的最近消息数。旧消息被摘要替换。",
       "memoryFlush": "压缩前记忆",

ui/web/src/pages/agents/agent-detail/config-sections/compaction-section.tsx

@@ -25,22 +25,11 @@ export function CompactionSection({ value, onChange }: CompactionSectionProps) {
           <Input
             type="number"
             step="0.05"
-            placeholder="0.75"
+            placeholder="0.85"
             value={value.maxHistoryShare ?? ""}
             onChange={(e) => onChange({ ...value, maxHistoryShare: numOrUndef(e.target.value) })}
           />
         </div>
-        <div className="space-y-2">
-          <InfoLabel tip={t(`${s}.minMessagesTip`)}>{t(`${s}.minMessages`)}</InfoLabel>
-          <Input
-            type="number"
-            placeholder="50"
-            value={value.minMessages ?? ""}
-            onChange={(e) => onChange({ ...value, minMessages: numOrUndef(e.target.value) })}
-          />
-        </div>
-      </div>
-      <div className="grid grid-cols-1 gap-4 sm:grid-cols-2">
         <div className="space-y-2">
           <InfoLabel tip={t(`${s}.keepLastMessagesTip`)}>{t(`${s}.keepLastMessages`)}</InfoLabel>
           <Input

ui/web/src/types/agent.ts

@@ -22,7 +22,6 @@ export interface SubagentsConfig {
 export interface CompactionConfig {
   reserveTokensFloor?: number;
   maxHistoryShare?: number;
-  minMessages?: number;
   keepLastMessages?: number;
   memoryFlush?: {
     enabled?: boolean;