@ai-agent-forge/agent-forge 0.86.7 → 0.88.0

package/dist/core/agent-session-compaction.js

@@ -7,7 +7,6 @@
  * spies keep working unchanged. Method-to-method calls go through the session
  * parameter (not direct function calls) to preserve that dispatch.
  */
-import { SUMMARIZATION_SYSTEM_PROMPT } from "@agent-forge/agent-core";
 import { contentText, retryAssistantCall, uuidv7 } from "@agent-forge/ai";
 import { completeSimple, streamSimple } from "@agent-forge/ai/compat";
 import { detachedSessionSnapshot, SessionCancelledError, } from "./agent-session-internal.js";
@@ -16,6 +15,7 @@ import { formatNoApiKeyFoundMessage, formatNoModelSelectedMessage } from "./auth
 import { calculateContextTokens, collectEntriesForBranchSummary, effectiveKeepRecentTokens, estimateContextTokens, estimateTokens, isTurnStartMessage, } from "./compaction/index.js";
 import { DEFAULT_COMPACTION_POLICY_NAME, SESSION_COMPACTION_COMMIT_EVENT_ID, SESSION_COMPACTION_COMMIT_EVENT_VERSION, } from "./extensions/index.js";
 import { SUMMARY_MODEL_INVOCATION_ERROR_CODES } from "./extensions/public-api.js";
+import { t } from "./i18n/index.js";
 import { getLatestCompactionEntry } from "./session-manager.js";
 function withoutDeletedHeaders(headers) {
     return headers
@@ -38,25 +38,6 @@ function estimateMessagesTokens(messages) {
 const OVERFLOW_TRIM_MAX_STAGES = 3;
 /** 溢出裁剪第 2 级的目标 token 预算（估算）。 */
 const OVERFLOW_TRIM_STAGE2_TARGET_TOKENS = 10_000;
-/**
- * microcompact 冷却长度（按模型请求计数）：relief 落地后置此值，之后每次
- * assistant message_end（无论 stopReason）减 1，归零前阈值 decide 以
- * `microcompactEnabled: false` 关闭便宜缓解，落回全量压缩；user message_start
- * 不复位——冷却按模型请求计数，与 turn 边界无关。
- *
- * 微观经济动机：每次 microcompact 触发都是一次中部历史改写，本网关前缀缓存
- * 不稳定（实测同请求命中 3-99%），改尾部命中 88-99.9%、改中部 40-50%——
- * 冷却把"阈值带上反复触发中部改写"的最坏情形封顶。
- */
-const MICROCOMPACT_COOLDOWN_ASSISTANT_RESPONSES = 3;
-/** 摘要自批判评审轮 prompt（gemini-cli 同款思路，compaction.summarySelfCritique 开启时启用）。 */
-function summarySelfCritiquePrompt(summary) {
-    return `Critically evaluate the summary in <summary> tags: it is the only context a continuation model will receive. Check for omissions that would break continuity - exact file paths, function/identifier names, error messages, user-stated constraints and preferences, decisions with rationale, and current task state. If something important is missing or wrong, output the corrected summary in the same format. If it is already complete, output it again verbatim. Output ONLY the summary.
-
-<summary>
-${summary}
-</summary>`;
-}
 /**
  * 溢出/截断裁剪失败文案：实际发生过裁剪时附上前缀（分语：context overflow
  * 维持既有精确文案，recoverable length 用截断响应措辞），0 次保持策略原文。
@@ -135,9 +116,7 @@ export async function getRequiredRequestAuth(session, model, signal) {
     }
     const isOAuth = session._modelRuntime.isUsingOAuth(model.provider);
     if (isOAuth) {
-        throw new Error(`Authentication failed for "${model.provider}". ` +
-            `Credentials may have expired or network is unavailable. ` +
-            `Run '/login ${model.provider}' to re-authenticate.`);
+        throw new Error(t("Authentication failed for \"{provider}\". Credentials may have expired or network is unavailable. Run '/login {provider}' to re-authenticate.", { provider: model.provider }));
     }
     throw new Error(formatNoApiKeyFoundMessage(model.provider));
 }
@@ -288,13 +267,6 @@ export async function compactBeforeNextAssistantResponse(session, context, signa
     if (!model || model.contextWindow <= 0) {
         return context;
     }
-    // Adaptive keepRecentTokens：小窗口把近期保护预算钳到可摘要窗口的一半
-    //（effectiveKeepRecentTokens，动机见其注释；大窗口原值不变），本路径与
-    // checkCompaction/compactInternal/runAutoCompactionInternal 一致使用 effective 值。
-    const effectiveSettings = {
-        ...settings,
-        keepRecentTokens: effectiveKeepRecentTokens(settings.keepRecentTokens, model.contextWindow, settings.reserveTokens),
-    };
     const contextTokens = estimateContextTokens(context.messages).tokens;
     const decision = await decideCompaction(session, {
         reason: "threshold",
@@ -306,11 +278,6 @@ export async function compactBeforeNextAssistantResponse(session, context, signa
             reserveTokens: settings.reserveTokens,
         },
         canContinue: false,
-        // 便宜缓解的事实输入与门控，与 checkCompaction 的 threshold decide 相同
-        //（设置开关 ∧ 冷却归零）：被任一条件挡住时策略不做便宜缓解，直接走全量压缩。
-        branchEntries: session.sessionManager.getBranch(),
-        keepRecentTokens: effectiveSettings.keepRecentTokens,
-        microcompactEnabled: settings.microcompactEnabled && session._microcompactCooldown === 0,
         signal: signal ?? session._policyDecisionAbortController?.signal,
     });
     await dispatchCompactionDecisionLifecycle(session, {
@@ -324,27 +291,6 @@ export async function compactBeforeNextAssistantResponse(session, context, signa
         reserveTokens: settings.reserveTokens,
         ...(decision.failureMessage === undefined ? {} : { failureMessage: decision.failureMessage }),
     });
-    // 策略便宜缓解（通用通道），镜像 checkCompaction 的 relief 落地分支：宿主原样
-    // 追加条目并重投影上下文，条目语义与投影归策略。执行上下文：本函数经默认
-    // prepareNextTurn loop link 在 agent 回合之间被 agent run await，不在
-    // _compactionQueue 内；手动压缩入口 compactInternal 先 _abortInternal() 等待
-    // 活跃回合结束，而本路径位于回合执行链上，回合结束前必先完成——relief 落地
-    // （同步 append + 投影）不会与手动压缩的摘要/提交交错，与 postAgentRun 的
-    // relief 落地同模式，不加队列锁。
-    if (decision.relief !== undefined && !decision.compact) {
-        session.sessionManager.appendCustomEntry(decision.relief.customType, decision.relief.data);
-        const sessionContext = session.sessionManager.buildSessionContext();
-        await session._applyContextTransition(sessionContext.messages, "compaction");
-        // 冷却起点：relief 已落地即开始计数（由 assistant message_end 递减）。
-        session._microcompactCooldown = MICROCOMPACT_COOLDOWN_ASSISTANT_RESPONSES;
-        session._hostLogger?.info("compaction", "policy relief applied", {
-            fields: { customType: decision.relief.customType, path: "pre-response" },
-        });
-        return {
-            ...context,
-            messages: session.agent.state.messages.slice(),
-        };
-    }
     if (!decision.compact)
         return context;
     await session._runAutoCompaction("threshold", false);
@@ -368,7 +314,7 @@ export async function runCompactionSummary(session, preparation, auth, customIns
     const runtime = session._capabilityRuntime;
     // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
     if (!runtime)
-        throw new Error("Capability runtime is not available");
+        throw new Error(t("Capability runtime is not available"));
     const result = await runtime.summarizeCompaction(session._compactionPolicyName, {
         preparation,
         model: summaryModelFacts(auth.model),
@@ -383,45 +329,14 @@ export async function runCompactionSummary(session, preparation, auth, customIns
         context: { version: 1, sessionId: session.sessionId, phase: "compaction" },
         execution,
     });
-    let summary = result.summary;
-    let usage = result.usage;
-    // 摘要自批判评审轮（settings compaction.summarySelfCritique，默认关闭）：
-    // 摘要是续接模型的全部上下文，漏一处路径/约束就断一环——多花一次小请求换
-    // 一遍"漏了什么"的自检（gemini-cli 两段式压缩的同款思路）。失败/中止/空文
-    // 本时保持第一轮摘要，评审轮不产生降级路径。
-    if (session.settingsManager.getCompactionSettings().summarySelfCritique && summary.trim() && !signal.aborted) {
-        const critique = await createSummaryModelInvocationV1(session, auth, summarizationRetryPolicy(session, execution.operationId, execution.inputRevision, signal), summarizationRetryCallbacks(session, { source: "compaction", reason }), { routingSessionId: session.sessionId }).invoke({
-            version: 1,
-            systemPrompt: SUMMARIZATION_SYSTEM_PROMPT,
-            prompt: summarySelfCritiquePrompt(summary),
-            maxTokens: 8192,
-            thinkingLevel: session.thinkingLevel === "off" ? "off" : "low",
-        }, signal);
-        if (critique.status === "completed" && critique.text && critique.text.trim()) {
-            summary = critique.text;
-            // 公共契约的 usage 是不透明透传；评审轮实际回传 provider Usage 形状，
-            // 这里按字段防御性收窄后并入压缩 usage。
-            const critiqueUsage = critique.usage;
-            if (usage && critiqueUsage) {
-                const add = (a, b) => a + (b ?? 0);
-                usage = {
-                    ...usage,
-                    input: add(usage.input, critiqueUsage.input),
-                    output: add(usage.output, critiqueUsage.output),
-                    cacheRead: add(usage.cacheRead ?? 0, critiqueUsage.cacheRead),
-                    cacheWrite: add(usage.cacheWrite ?? 0, critiqueUsage.cacheWrite),
-                };
-            }
-        }
-    }
     // 包内收窄：invocation 结果按 provider 原样透传 usage；压缩 usage 合并需要内部 Usage 形状。
-    return { ...result, summary, usage };
+    return { ...result, usage: result.usage };
 }
 export async function decideCompaction(session, input) {
     const runtime = session._capabilityRuntime;
     // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
     if (!runtime)
-        throw new Error("Capability runtime is not available");
+        throw new Error(t("Capability runtime is not available"));
     // 内核机制兜底：默认策略名且无人注册（默认插件被禁用或失败隔离）时，机制层
     // 不做压缩决策（无执行者），回合继续；缺失经 warn 日志与手动 /compact 的显式
     // 错误暴露。默认插件不得成为隐藏前提（宪法 §2/§5）。显式配置的其他策略名缺
@@ -434,7 +349,6 @@ export async function decideCompaction(session, input) {
         return { compact: false, retry: false };
     }
     return runtime.decideCompaction(session._compactionPolicyName, {
-        enabled: session.settingsManager.getCompactionSettings().enabled,
         recoveryAttempted: session._overflowRecoveryAttempted,
         ...input,
         context: { version: 1, sessionId: session.sessionId, phase: "compaction" },
@@ -446,7 +360,7 @@ export async function prepareCompaction(session, pathEntries, settings, executio
     const runtime = session._capabilityRuntime;
     // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
     if (!runtime)
-        throw new Error("Capability runtime is not available");
+        throw new Error(t("Capability runtime is not available"));
     const preparation = await runtime.prepareCompaction(session._compactionPolicyName, {
         entries: pathEntries,
         settings,
@@ -466,7 +380,9 @@ export async function prepareCompaction(session, pathEntries, settings, executio
         !Number.isFinite(value.tokensBefore) ||
         typeof value.fileOps !== "object" ||
         typeof value.settings !== "object") {
-        throw new Error(`Compaction policy returned an invalid preparation: ${session._compactionPolicyName}`);
+        throw new Error(t("Compaction policy returned an invalid preparation: {policyName}", {
+            policyName: session._compactionPolicyName,
+        }));
     }
     return preparation;
 }
@@ -490,7 +406,7 @@ export function enqueueCompaction(session, operation) {
     const queued = previous
         .then(() => {
         if (session._reloadInProgress) {
-            throw new Error("Session reload in progress; compaction was not started");
+            throw new Error(t("Session reload in progress; compaction was not started"));
         }
         return operation();
     })
@@ -525,11 +441,16 @@ export function commitCompaction(session, input) {
         usage: input.usage,
     });
     if (commit.status === "revision_conflict") {
-        throw new Error(`Compaction commit revision conflict: expected ${input.expectedRevision.sequence}/${input.expectedRevision.leafId ?? "root"}, current ${commit.currentRevision.sequence}/${commit.currentRevision.leafId ?? "root"}`);
+        throw new Error(t("Compaction commit revision conflict: expected {expectedSequence}/{expectedLeafId}, current {currentSequence}/{currentLeafId}", {
+            expectedSequence: input.expectedRevision.sequence,
+            expectedLeafId: input.expectedRevision.leafId ?? "root",
+            currentSequence: commit.currentRevision.sequence,
+            currentLeafId: commit.currentRevision.leafId ?? "root",
+        }));
     }
     const entry = session.sessionManager.getEntry(commit.entryId);
     if (entry?.type !== "compaction" || entry.operationId !== input.operationId) {
-        throw new Error(`Compaction commit entry is unavailable: ${commit.entryId}`);
+        throw new Error(t("Compaction commit entry is unavailable: {entryId}", { entryId: commit.entryId }));
     }
     return { entry, status: commit.status, revision: commit.revision };
 }
@@ -539,7 +460,7 @@ export async function dispatchCompactionCommitLifecycle(session, reason, commit)
         const runtime = session._capabilityRuntime;
         // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
         if (!runtime)
-            throw new Error("Capability runtime is not available");
+            throw new Error(t("Capability runtime is not available"));
         const entry = commit.entry;
         const data = {
             version: 1,
@@ -584,7 +505,7 @@ export async function decideOverflow(session, message, execution, signal) {
     const runtime = session._capabilityRuntime;
     // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
     if (!runtime)
-        throw new Error("Capability runtime is not available");
+        throw new Error(t("Capability runtime is not available"));
     return runtime.decideOverflow(session._overflowPolicyName, {
         message,
         contextWindow: session.model?.contextWindow,
@@ -634,9 +555,9 @@ export async function compactInternal(session, customInstructions) {
             // Check why we can't compact
             const lastEntry = pathEntries[pathEntries.length - 1];
             if (lastEntry?.type === "compaction") {
-                throw new Error("Already compacted");
+                throw new Error(t("Already compacted"));
             }
-            throw new Error("Nothing to compact (session too small)");
+            throw new Error(t("Nothing to compact (session too small)"));
         }
         // Shared default summary generator, also used by automatic compaction.
         const { summary, firstKeptEntryId, tokensBefore, usage, details } = await runCompactionSummary(session, preparation, auth, customInstructions, session._compactionAbortController.signal, "manual", execution);
@@ -746,7 +667,7 @@ export async function checkCompaction(session, assistantMessage, skipAbortedChec
     const settings = session.settingsManager.getCompactionSettings(session.model ?? undefined);
     // Adaptive keepRecentTokens（effectiveKeepRecentTokens，动机见其注释：座席实测
     // 30K 窗口 20K 保护预算使可摘要前缀为零，阈值反复命中却无事可做）。一次性
-    // 替换后全文一致使用：下方 trim 阶梯 stage 0 目标与 threshold decide 请求。
+    // 替换后全文一致使用：下方 trim 阶梯 stage 0 目标。
     const effectiveSettings = {
         ...settings,
         keepRecentTokens: effectiveKeepRecentTokens(settings.keepRecentTokens, session.model?.contextWindow ?? 0, settings.reserveTokens),
@@ -954,13 +875,6 @@ export async function checkCompaction(session, assistantMessage, skipAbortedChec
             reserveTokens: settings.reserveTokens,
         },
         canContinue: false,
-        // 便宜缓解的事实输入：策略可以据此先做零成本的本地上下文缓解（如
-        // bootstrap 策略的 microcompact），把 LLM 摘要留作兜底。分支快照仅为
-        // 策略读取，内核不解释其内容。门控（设置开关 ∧ 冷却归零）被挡住时，
-        // 策略直接走全量压缩。
-        branchEntries: session.sessionManager.getBranch(),
-        keepRecentTokens: effectiveSettings.keepRecentTokens,
-        microcompactEnabled: effectiveSettings.microcompactEnabled && session._microcompactCooldown === 0,
         signal,
     });
     await dispatchCompactionDecisionLifecycle(session, {
@@ -972,21 +886,6 @@ export async function checkCompaction(session, assistantMessage, skipAbortedChec
         reserveTokens: settings.reserveTokens,
         ...(decision.failureMessage === undefined ? {} : { failureMessage: decision.failureMessage }),
     });
-    // 策略便宜缓解（通用通道）：策略通过 decision.relief 返回一个自定义条目，
-    // 宿主原样追加并重投影上下文——内核不解释条目语义，投影由策略经公共
-    // registerEntryProjection 通道自行注册（宪法 §4/§5：策略在插件，机制在内核）。
-    if (decision.relief !== undefined && !decision.compact) {
-        session.sessionManager.appendCustomEntry(decision.relief.customType, decision.relief.data);
-        const sessionContext = session.sessionManager.buildSessionContext();
-        await session._applyContextTransition(sessionContext.messages, "compaction");
-        // 冷却起点：relief 已落地即开始计数（本次改写后的下一个模型请求起算，
-        // 由 assistant message_end 递减，见 AgentSession）。
-        session._microcompactCooldown = MICROCOMPACT_COOLDOWN_ASSISTANT_RESPONSES;
-        session._hostLogger?.info("compaction", "policy relief applied", {
-            fields: { customType: decision.relief.customType },
-        });
-        return session.agent.hasQueuedMessages();
-    }
     return decision.compact ? session._runAutoCompaction("threshold", decision.retry === true) : false;
 }
 /**
@@ -1197,13 +1096,17 @@ export function summarizationRetryCallbacks(session, source) {
         },
     };
 }
-/** Adapt the selected capability retry policy to the shared summary executor. */
+/**
+ * Adapt the selected capability retry policy to the shared summary executor.
+ * `...settings` 展开携带 `enabled: true`（pi-ai `RetryPolicy` 形状要求，
+ * settings-manager 聚合恒填 true——产品侧自动重试无条件生效，D-079）。
+ */
 export function summarizationRetryPolicy(session, operationId, inputRevision, signal) {
     const settings = session.settingsManager.getRetrySettings();
     const runtime = session._capabilityRuntime;
     // v8 ignore next -- capability runtime 恒由 _buildRuntime 创建，dispose 后不再调用，运行期恒存在
     if (!runtime)
-        throw new Error("Capability runtime is not available");
+        throw new Error(t("Capability runtime is not available"));
     const execution = executionFacts(session, "retry", operationId ? `${operationId}:retry` : undefined, inputRevision);
     return {
         ...settings,
@@ -1215,7 +1118,6 @@ export function summarizationRetryPolicy(session, operationId, inputRevision, si
         }),
         decide: ({ message, retryable, attempt, maxAttempts, baseDelayMs }) => runtime.decideRetry(session._retryPolicyName, {
             retryable,
-            enabled: settings.enabled,
             attempt,
             maxAttempts,
             baseDelayMs,
@@ -1240,11 +1142,11 @@ export function summarizationRetryPolicy(session, operationId, inputRevision, si
  */
 export async function navigateTree(session, targetId, options = {}) {
     if (session.isStreaming) {
-        throw new Error("Wait for the current response to finish before navigating the session tree.");
+        throw new Error(t("Wait for the current response to finish before navigating the session tree."));
     }
     // Compaction mutates the active branch; navigating mid-compaction would race the summary commit.
     if (session._compactionAbortController !== undefined || session._autoCompactionAbortController !== undefined) {
-        throw new Error("Wait for the current compaction or tree navigation to finish before navigating the session tree.");
+        throw new Error(t("Wait for the current compaction or tree navigation to finish before navigating the session tree."));
     }
     const oldLeafId = session.sessionManager.getLeafId();
     // No-op if already at target
@@ -1253,11 +1155,11 @@ export async function navigateTree(session, targetId, options = {}) {
     }
     // Model required for summarization
     if (options.summarize && !session.model) {
-        throw new Error("No model available for summarization");
+        throw new Error(t("No model available for summarization"));
     }
     const targetEntry = session.sessionManager.getEntry(targetId);
     if (!targetEntry) {
-        throw new Error(`Entry ${targetId} not found`);
+        throw new Error(t("Entry {id} not found", { id: targetId }));
     }
     // Collect entries to summarize (from old leaf to common ancestor)
     const { entries: entriesToSummarize } = collectEntriesForBranchSummary(session.sessionManager, oldLeafId, targetId);
@@ -1265,10 +1167,10 @@ export async function navigateTree(session, targetId, options = {}) {
     const replaceInstructions = options.replaceInstructions;
     const label = options.label;
     if (session._reloadInProgress) {
-        throw new Error("Session reload in progress; tree navigation is unavailable");
+        throw new Error(t("Session reload in progress; tree navigation is unavailable"));
     }
     if (session._branchSummaryAbortController) {
-        throw new Error("Branch navigation is already in progress");
+        throw new Error(t("Branch navigation is already in progress"));
     }
     // Set up abort controller for summarization
     session._branchSummaryAbortController = new AbortController();
@@ -1287,7 +1189,7 @@ export async function navigateTree(session, targetId, options = {}) {
             const branchSummarySettings = session.settingsManager.getBranchSummarySettings();
             const runtime = session._capabilityRuntime;
             if (!runtime || !runtime.hasBranchSummaryPolicy(session._branchSummaryPolicyName)) {
-                throw new Error(`Branch summary policy is not registered: ${session._branchSummaryPolicyName}`);
+                throw new Error(t("Branch summary policy is not registered: {name}", { name: session._branchSummaryPolicyName }));
             }
             const execution = executionFacts(session, "branch_summary", `branch_summary:${session.sessionId}:${oldLeafId ?? "root"}:${targetId}`);
             const result = await runtime.summarizeBranch(session._branchSummaryPolicyName, {