@lingxiao-office/sdk 1.0.0

package/dist/agents/prompts/i18n/catalog.js

@@ -0,0 +1,498 @@
+import { getLanguage, getSessionLanguage, normalizeLanguage, } from '../../../i18n.js';
+const catalogByLocale = {
+    zh: {
+        locale: 'zh',
+        llmLanguageDirective: '请使用中文回复。保留 XML tags、JSON keys、tool names、enum values、Context Manifest、graph_contract、lingxiao_completion 等机器协议字段的英文原文；这些协议标识保持原样。',
+        leader: {
+            availableSkillsHeading: '当前可用 Skills 列表',
+            availableSkillsIntro: '以下 Skills 可在创建角色（define_agent_role）或任务（create_task）时通过 skill_names 参数指定，注入 Worker 的执行上下文：',
+            eternalGoal: {
+                evidenceRequired: '必须提供 evidence（完成证据）才能标记目标为已完成。',
+                summaryRequired: '必须提供 summary（完成总结）才能标记目标为已完成。',
+                noActiveGoal: '当前没有活跃的 Eternal 目标，无法标记完成。',
+                goalCompleted: (description) => `Eternal 目标已完成：${description}`,
+                evidenceLabel: '完成证据',
+                completedStatus: 'Eternal · 目标已完成，待命中',
+                completedToolResult: (description) => `Eternal 目标「${description}」已标记为完成。已自动切回 manual 模式，Leader 进入待命。如需继续可设置新 goal。`,
+                patrolDeliveryStep: '5. **收尾交付**：目标完成时调用 complete_eternal_goal(evidence, summary) 标记完成，不要转去无关优化',
+                patrolCompletionGuidanceEarly: '如果目标确实已完成（有测试通过、构建成功、代码提交等具体证据），调用 complete_eternal_goal 标记完成；否则继续找下一步。',
+                patrolCompletionGuidanceLate: '除非目标已有明确完成证据（此时调用 complete_eternal_goal），否则不要因为本轮空闲而转去无关优化。',
+                statusStandby: 'Eternal · 待命中',
+                toolDescription: '标记当前 Eternal 目标为已完成。仅当有充分证据证明目标已达成时调用。调用后 goal 被清除，自动切回 manual 模式，Leader 进入 idle 待命。如需继续可设置新 goal。不要在目标仍有未验证部分时调用。',
+                toolEvidenceDescription: '目标完成的具体证据（测试通过、构建成功、代码提交等可验证事实）',
+                toolSummaryDescription: '目标完成总结，包含完成了什么、关键产出和后续建议',
+            },
+        },
+        workerTask: {
+            taskContextHeading: '任务上下文',
+            taskGoalLabel: '目标',
+            taskDescriptionLabel: '描述',
+            workspaceLabel: '工作区',
+            writeScopeLabel: '写入范围',
+            sessionLabel: '会话',
+            budgetFast: '**预算**: 快速任务，聚焦直接证据和最小改动。',
+            budgetMedium: '**预算**: 中等任务，先理解再动手，完成后验证。',
+            budgetLarge: '**预算**: 大型任务，先规划再执行；复杂度高时写 Scratchpad 跟踪。',
+            contextManifestTitle: '### Context Manifest（系统统一注入）',
+            leaderContextHeading: 'Leader 背景上下文（启动前必读）',
+            leaderContextIntro: '由 Leader 在分配任务时提供，包含用户意图、已知信息、关键决策：',
+            injectedSkillBodiesTitle: 'Injected Skill Bodies',
+            knowledgeGraphHeading: '知识图谱协议',
+            knowledgeGraphIntro: '共享知识图谱是跨 Agent 的实时公共记忆。先阅读 Context Manifest 中的黑板快照；其中 Facts/Intents 会包含 content、tags、evidence 和关系。',
+            knowledgeGraphAction: '执行时沿已有事实继续推进；发现新事实、新方向或矛盾时通过结构化代码块补充公共记忆：',
+            skillPathRulesHeading: 'Skill 路径规则',
+            skillPriorityLine: ({ projectSkillsDir, globalSkillsDir }) => `优先级: 项目级 > 插件贡献 > 用户级 > 内置。项目级目录: ${projectSkillsDir}/ | 用户级目录: ${globalSkillsDir}/ | 插件/内置目录以 Context Manifest 的 skills 路径为准`,
+            skillPathLine: 'Skill 文件路径以注入区 <skill path="..."> 的真实 path 字段为准；需要完整技能时读取该 path。',
+            autoInjectedLine: (names) => `自动注入: ${names.join(', ')}`,
+            skillBodiesLine: '技能正文和插件来源已进入 Context Manifest；若 truncated=true，必须按真实 path 读取完整技能或相对引用文档后再做关键办公产物。',
+            deliverySopHeading: '交付完整性与验收 SOP',
+            deliveryScopeRule: '发现任务描述只覆盖子集、但用户目标明显是完整项目/完整功能时，先通过 send_message 升级 Leader 对齐范围，并按完整目标继续规划',
+            deliveryCompletenessRule: '完成标准包含可用性、集成链路、错误/空/加载态和可追溯验证证据；骨架、占位页面、假数据或未接线功能仅在任务契约明确要求时作为对应阶段产物',
+            deliveryPageEvidenceRule: '页面级任务同时提供工程证据和真实页面证据：`npm test`、类型检查或构建成功作为工程证据，浏览器验收作为用户可见路径证据',
+            collaborationHeading: '协作与守卫',
+            collaborationRules: ({ taskId }) => [
+                '前序 Agent 笔记、产物、契约和文件树在 Context Manifest 中；开工先读 manifest，随后用 `read_work_notes` 或 manifest 指向的 artifact 补齐证据',
+                `达到验收标准后调用 \`attempt_completion\` 收尾；每个任务都必须提供 \`contract_compliance\` 契约遵守证明（surface/status/evidence/deviations），无跨栈契约时 surface 使用 \`task:${taskId}\``,
+                'coding 最终摘要必须含文件路径和验证证据；缺少契约遵守证明时系统会拒绝 complete',
+                '代码修改后必须验证：编译（`tsc --noEmit`）或测试，确认无报错',
+            ],
+            teamCommunicationHeading: 'Team 通信（如果你属于 team）',
+            teamCommunicationRules: [
+                '启动任务后先调用 `team_inbox(unread_only=true, mark_read=true)`，直接处理已有 P2P 私信和 team 广播，并把同伴消息纳入当前判断',
+                '开工前必须阅读 Context Manifest 的黑板快照，优先查找 `[contract]` / `[design_doc]` 节点；跨栈实现按现有 contract 执行，缺失或冲突时先通知 team/Leader 并提交 v+1 契约',
+                '需要确认团队成员时调用 `team_manage(action="list_members")`；P2P 目标必须是 `interactive=true` 成员：`team_message(target_type="member", target="成员名", content=...)`。`not_dispatched` 成员走 team 广播，或用 `send_message` 请求 Leader 先 dispatch',
+                '想了解同伴进度（谁在做什么、哪些任务就绪/阻塞）时调用 `team_manage(action="task_board")`，读取任务板后主动对齐写入范围、接口契约或集成窗口',
+                '需要通知全队时调用 `team_message(target_type="team", target="team名", content=...)`；广播用于接口契约、全局约定、跨模块风险、集成窗口等全队相关事项；目标字段始终使用 `target_type + target`',
+                '你的产出会影响其他成员时必须主动通知：API/数据结构/文件路径/组件 props/环境变量/运行命令/验收口径发生新增或变化，都要发 P2P 或广播',
+                '需要对方明确答复时用完整调用 `team_message(target_type="member", target="<receiver>", content="...", type="request", request_id="<surface>@v<N>")`；收到 request 并处理后由你本人回给原发送者：`team_message(target_type="member", target="<original sender>", content="...", type="ack", request_id="<same>")`。ack 由处理 request 的成员发出。`team_inbox` 的 `ack_status` 会提示你「还在等谁的 ack」和「你欠谁一个 ack」',
+                '收到 `[协商请求]` 时说明你和某成员存在文件/资源冲突：先直接与对方协商收敛（拆分写入路径或串行化），达成一致后按指令里的 request_id 回 ack，且必须填写 target_type/target/content；谈不拢再升级 Leader',
+                '跨栈接口/schema/props/env/命令契约发生新增或变化时，最终输出必须包含 `graph_contract` 代码块：字段至少包含 `surface`、`title`、`content`；重要方案沉淀用 `graph_design_doc` 代码块',
+                '遇到依赖阻塞时先直接联系对应成员；需要用户/Leader 决策、跨 team 协调或等待超时时升级给 Leader',
+                '结束前再调用一次 `team_inbox`，确认没有未处理协作消息和未闭环的 ack；收尾时用 `write_work_note` 汇总已发送/已接收的关键协作信息，并在最终结果摘要里说明',
+                '战略级汇报（任务收尾、需要 Leader 决策）走 `send_message(recipient="leader", ...)`；任务完成时直接回复最终结果摘要，系统会自动把它作为完成报告回传给 Leader',
+            ],
+            startInstruction: '开始执行。完成后直接回复结果摘要；阻塞时用 send_message 联系 Leader。',
+        },
+        sharedFragments: {
+            capabilitySurfaceHeading: '能力面协议（Plugin / Skill / MCP）',
+            capabilitySurfaceRules: [
+                'Plugin 是本地能力包和分发单元；启用后把 skills、MCP servers、apps/assets、tools/hooks/scripts 贡献到凌霄能力面',
+                'Skill 是注入 Agent 上下文的执行知识、流程和领域约束；来源优先级为项目级 > 插件贡献 > 用户级 > 内置，Leader 在 define_agent_role/create_task 中用 skill_names 绑定',
+                'MCP 是连接外部系统的 tools/resources/prompts 协议；运行时统一通过 `mcp(action="list_servers|list_tools|call_tool|list_resources|read_resource", ...)` 发现和访问 tools/resources',
+                '需要 prompts、resource templates 或服务能力快照时，使用同一 `mcp` 工具的 action="list_prompts|get_prompt|list_resource_templates|capability_snapshot"',
+                '插件贡献的 MCP server 会同步到 settings.mcp.servers；所有 MCP server 通过同一个 `mcp` 工具入口访问',
+                '插件 tools/hooks/scripts 只有接入 ToolRegistry 或 MCP server 后才属于 runtime 可调用能力；其余情况作为 marketplace、packaging、hook metadata 和本地源码线索进入 Context Manifest',
+                '需要完整 Skill 时读取注入区 `<skill path="...">` 的真实路径；调用 MCP 前先 list_servers/list_tools/list_resources，再按真实 schema 调用',
+                '能力调用产生的文件、命令、MCP resource URI、URL、截图/报告路径和验证结果写入 write_work_note、attempt_completion.evidence_refs 或 lingxiao_completion.evidence_refs，让 Context Manifest 继续传递给后续 Agent',
+            ],
+            externalCompletionHeading: '凌霄完成报告协议',
+            externalCompletionIntro: '最终答复先写给 Leader 看的自然语言摘要，末尾附加一个 ```lingxiao_completion 代码块，Lingxiao 会解析它并注入后续 Agent 上下文。',
+            externalCompletionSummaryExample: '一句话说明完成了什么',
+            externalCompletionNoDeviationExample: '无',
+            externalCompletionNextStepExample: '可选的后续建议',
+            externalCompletionNotes: '路径使用工作区相对路径或绝对路径；contract_compliance 是必填契约遵守证明，包含 surface/status/evidence/deviations；evidence_refs 写 MCP resource URI、URL、截图/报告路径或外部工具结果 ID；blocked_by_discovery 写新发现但未解决的依赖，needs_leader_coordination=true 表示需要 Leader 协调；验证写真实命令与结果；跨栈接口/schema/props/env/命令契约变化同时输出 graph_contract 代码块。',
+            externalCompletionBrowserEvidence: '前端/全栈任务完成报告包含真实浏览器验收证据；构建、lint 或单元测试作为工程证据，页面级验收另列真实页面证据。浏览器无法启动时写入 blocked/skipped 与错误原因；视觉或交互 passed 结论只在真实页面证据齐全时给出。',
+        },
+        tools: {
+            registry: {
+                schemaValidationFailed: 'schema 校验失败',
+                deferredToolDescription: (name) => `延迟加载工具：${name}`,
+                argumentValidationMessage: (formatted) => `参数校验失败：${formatted}`,
+                argumentValidationFix: '按工具 schema 修正参数；仅传 schema 声明字段，并为互斥目标选择一种表达。可选字段未知时直接省略，不要传空字符串。',
+                toolNotFoundMessage: (name) => `未找到工具：${name}`,
+                toolNotFoundCause: (name) => `没有注册名为 ${name} 的工具。`,
+                toolNotFoundFix: (candidates) => candidates.length > 0 ? `请改用候选工具之一：${candidates.join(', ')}` : '请检查工具名或先加载/注册该工具。',
+                argumentParseFix: '工具参数必须直接传 JSON object/array 值；不要把参数包成字符串。',
+                scopeForbiddenMessage: (name) => `当前上下文不可用工具：${name}`,
+                scopeForbiddenCause: (name) => `工具 ${name} 仅允许 Leader 上下文调用。`,
+                scopeForbiddenFix: '不要从 Worker 直接调用 Leader 元工具；需要编排时向 Leader 汇报或使用任务允许的 worker 工具。',
+                validationFailedMessage: '参数校验失败。',
+                validationFailedFix: '按工具 schema 修正参数。',
+                fileMustReadFirstMessage: (path) => `编辑前必须先读取文件：请先使用 file_read 读取 "${path}"，确认文件内容后再进行编辑。`,
+                fileMustReadFirstCause: '为避免覆盖未知用户改动，编辑/覆盖已有文件前必须先读文件。',
+                fileMustReadFirstFix: '先调用 file_read 读取该 path，再重试当前编辑。',
+                officeModeRequiredMessage: (name) => `办公工具需要 Office 模式：${name}`,
+                officeModeRequiredCause: '当前会话未开启 Office 模式，办公工具不会执行。',
+                officeModeRequiredFix: '请先使用 /office on 或前端 Office 开关开启当前会话的 Office 模式，再重试。',
+                permissionRequiredFix: '等待用户授权，或选择无需该权限的替代工具/方案。',
+            },
+            structuredPatch: {
+                reuseOriginalReplace: '<在这里复用原始 replace 文本>',
+                reuseOriginalContent: '<在这里复用原始 content 文本>',
+                wholeLineRangeNote: '仅在确实要替换列出的整段行范围时使用。',
+                exactOccurrenceNote: '仅在确实要替换这一处精确匹配时使用。',
+                finalOccurrenceNote: '仅在确实要替换最后一处匹配时使用。',
+                insertExactOccurrenceNote: '仅在确实要插入到这一处精确匹配之后时使用。',
+                insertFinalOccurrenceNote: '仅在确实要插入到最后一处匹配之后时使用。',
+                insertWholeRangeNote: '仅在确实要插入到整段匹配行/范围之后时使用。',
+                retryRuleSingleShape: '重试时只传 {path,hunks}；每个 hunk 必须只保留一种合法形态。不要把 received_fields 原样复制回 retry args。',
+                ambiguousSearchFix: '优先使用 retry_args.first_occurrence 或 retry_args.last_occurrence 指定目标；确认要全部替换时才使用 retry_args.replace_all。若只是追加长文档，直接用 retry_args.append_eof。',
+            },
+            astQuery: {
+                actionDescription: '基于 AST 的查询动作。',
+                symbolDescription: 'definitions、references、call_graph 或 implementors 使用的符号名。',
+                fileDescription: '可选的项目相对路径过滤，例如 src/index.ts。',
+                namePatternDescription: 'action=pattern 时应用到 AST 声明名称的正则表达式。',
+                kindsDescription: 'action=pattern 使用的声明类型。',
+                maxDepthDescription: 'action=call_graph 的最大调用图深度。',
+                limitDescription: '最大结果数量。',
+                symbolRequiredMessage: (action) => `ast_query action "${action}" 需要非空 symbol。`,
+                symbolRequiredFix: (action) => `使用 {"action":"${action}","symbol":"TargetName"} 重试。`,
+            },
+        },
+        judges: {
+            nextSpeaker: {
+                toolDescription: '返回 eternal mode 中 assistant response 在 stop finish reason 后是否应立即继续。',
+                system: [
+                    '你负责判断 eternal mode 中的自主 Leader 在模型返回 stop/end_turn finish reason 后是否应该立即继续。',
+                    '使用语义判断，不要使用本地文本模式规则。',
+                    '只有当可见 assistant 输出实质未完成、明显开始了未收尾的结构，或明确显示正处于思路中途且必须在同一轮继续时，才继续。',
+                    '不要仅因为可能存在未来任务而继续；open work 由 runtime 单独处理。',
+                    '只调用 submit_next_speaker_verdict 一次作为回复。',
+                ].join(' '),
+            },
+            eternalPatrol: {
+                toolDescription: '决定 eternal-mode patrol 现在应运行 LLM patrol、静默跳过，还是把控制权交还给用户。',
+                actionDescription: 'patrol: 项目状态有值得调查的新信号；skip: 噪音或已处理变更，不消耗 LLM tokens；yield_user: 没有有意义的剩余工作，把控制权交还给人类。',
+                reasonDescription: '简短理由（一句话）。',
+                system: [
+                    '你是 Eternal Mode 自主编排器的 patrol judge。',
+                    '判断 Leader 是否应消耗 tokens 运行另一轮 patrol、静默跳过，或把控制权交还给用户。',
+                    '当 eternal_goal 存在时，这是 Goal Mode：保护该目标作为最高优先级，直到有具体证据表明目标已完成，或没有仅靠人类输入就无法继续。',
+                    '当项目 fingerprint 发生变化，且存在具体失败、新任务、新 blackboard contract 或未回复的 teammate message 时，优先选择 patrol。',
+                    '存在 active eternal_goal 时，如果目标仍需要规划、创建任务、派发、实现、验证或最终交付，即使项目 fingerprint 未变化，也优先选择 patrol。',
+                    '当 fingerprint 变化但 diff 看起来像 Leader 自己上一轮输出、heartbeat 或外观噪音时，优先选择 skip。',
+                    '当没有 eternal_goal、fingerprint 未变化、last_patrol_outcome != productive 且 has_open_work=false 时，优先选择 yield_user。重复泛化 idle patrol 是浪费。',
+                    '不要仅因为一轮 idle 或 fingerprint 未变化，就对 active eternal_goal 返回 yield_user。',
+                    '只调用 submit_eternal_verdict 一次作为回复。',
+                ].join(' '),
+            },
+            workerCompletion: {
+                toolDescription: '返回 worker final text 是否是有效的任务完成。',
+                reasonDescription: '用于 verdict 的简短稳定 reason。',
+                feedbackDescription: '如果拒绝完成，给 worker 的具体反馈；如果接受则为空字符串。',
+                system: [
+                    '你是自主软件工程 worker 的 completion judge。',
+                    '使用语义判断，不要做关键词匹配。',
+                    '当 worker 已产出与任务相关的有意义输出时应接受，即使输出是部分交付、摘要或报告，且没有覆盖每个请求细节。',
+                    '只有当文本明显仍在规划未来工作且没有做任何事，或隐藏了未解决的执行失败时，才拒绝。',
+                    '保持宽松：如果 worker 明确取得进展并产出实质内容，即使并非各方面都完美或完整，也应接受。',
+                    '只调用 submit_completion_verdict 一次作为回复。',
+                ].join(' '),
+            },
+            scratchpadReview: {
+                toolDescription: '返回 scratchpad 是否仍包含有意义的未完成 follow-up 工作。',
+                system: [
+                    '你负责审查 session scratchpad notes 中未完成的尾部工作。',
+                    '使用语义判断，不要匹配 TODO 关键词。',
+                    '只有当笔记仍暗示真实待处理的验证、交付、清理或阻塞工作时，才标记 follow-ups。',
+                    '如果所有 notes 都已完成，返回 has_follow_ups=false。',
+                    '只调用 submit_scratchpad_followup_review 一次作为回复。',
+                ].join(' '),
+            },
+            workflowCondition: {
+                taskInstructions: [
+                    '使用下面的 input 评估 workflow condition。',
+                    '只返回以下精确 JSON 值之一: true 或 false。',
+                    '不要调用工具。不要解释。',
+                ],
+            },
+            taskClassification: {
+                toolDescription: '对最匹配 worker assignment 的 task prompt mode 进行分类。',
+                system: [
+                    '你负责为软件编排运行时分类 worker task prompt mode。',
+                    '综合 assignment 语义、task context 和 graph digest 进行判断。',
+                    '初始框定或目标发现工作选择 bootstrap。',
+                    '局势分析、缺口综合或规划下一步探索方向选择 reason。',
+                    '执行具体 intent、调查、验证或实现路径选择 explore。',
+                    '普通编码、研究、审查、QA 或交付工作，且不需要 blackboard-specific prompt mode 时选择 generic。',
+                    '只调用 submit_task_classification 一次作为回复。',
+                ].join(' '),
+            },
+            intentClassification: {
+                toolDescription: '对用户消息的意图进行分类。',
+                system: [
+                    'Leader 在同一次主 LLM 调用中负责记录用户消息意图；不要额外发起分类 LLM 调用。',
+                    '必须根据整句语义判断真实意图，不要使用关键词匹配、正则匹配或局部词命中。',
+                    'record_capability_intent 记录的是 capability envelope，不是单标签；primaryIntent 只是摘要，grants/denies/requiredGates/constraints 才是 gate 依据。',
+                    '根据完整语义填写 primaryIntent、scope、phase、grants、denies、requiredGates、constraints；不要用关键词匹配。',
+                    'grants/denies 只允许五类粗能力：read/write/shell/task/dispatch。read=读/搜索/分析/计划；write=写 workspace 文件；shell=命令/git/npm/test/deploy/python/terminal；task=任务图；dispatch=派发 worker。',
+                    '只读/解释/方案类请求默认 scope=read_only，只授予 read，并用 denies 禁止 write/shell/task/dispatch。',
+                    '实现/修复类请求按用户授权授予 read/write；如果用户说不要命令/不要 git/不要 deploy/不要 npm/test，一律 deny shell；不要派 worker 则 deny dispatch。',
+                    '完整项目/复杂项目应表达为 implement + project/workspace scope + design/prepare phase + read/write/task/dispatch grants，并按需要加入 blueprint_coverage/verify_after_change gate。',
+                    '每个新的用户 turn 如看到 record_capability_intent 可用，应先调用一次记录 profile；如果该工具不可用或工具结果提示本轮已记录，绝不要重复调用，直接继续执行用户请求。',
+                ].join(' '),
+            },
+            toolSelection: {
+                toolDescription: '选择应暴露给下一次 agent LLM call 的 tool names。',
+                selectedToolNamesDescription: '当前任务和角色的工具名，按优先级顺序排列。',
+                system: [
+                    '你为自主软件工程 agent 选择工具。',
+                    '使用当前 role、task、workspace scope 和 recent conversation，选择支持下一次 LLM call 的工具。',
+                    '按优先级顺序返回工具名。当规划、执行、验证、协作和完成能力适合任务时，包含足够的相关工具。',
+                    'runtime 会在你选择之后强制执行 token budget。',
+                    '只调用 submit_tool_selection 一次作为回复。',
+                ].join(' '),
+            },
+        },
+    },
+    en: {
+        locale: 'en',
+        llmLanguageDirective: 'Please respond in English. Preserve machine protocol identifiers exactly as written, including XML tags, JSON keys, tool names, enum values, Context Manifest, graph_contract, and lingxiao_completion.',
+        leader: {
+            availableSkillsHeading: 'Currently Available Skills',
+            availableSkillsIntro: 'The following Skills can be specified through the skill_names parameter when creating roles (define_agent_role) or tasks (create_task), and will be injected into the Worker execution context:',
+            eternalGoal: {
+                evidenceRequired: 'The evidence parameter is required to mark the goal as complete.',
+                summaryRequired: 'The summary parameter is required to mark the goal as complete.',
+                noActiveGoal: 'No active Eternal goal exists. Cannot mark as complete.',
+                goalCompleted: (description) => `Eternal goal completed: ${description}`,
+                evidenceLabel: 'Evidence',
+                completedStatus: 'Eternal · Goal completed, standing by',
+                completedToolResult: (description) => `Eternal goal "${description}" marked as complete. Auto-switched to manual mode. Leader is standing by — set a new goal to continue.`,
+                patrolDeliveryStep: '5. **Deliver**: When the goal is complete, call complete_eternal_goal(evidence, summary) to mark it done. Do not drift to unrelated optimizations.',
+                patrolCompletionGuidanceEarly: 'If the goal is truly complete (with concrete evidence such as passing tests, successful builds, or committed code), call complete_eternal_goal to mark it done; otherwise continue to the next step.',
+                patrolCompletionGuidanceLate: 'Unless the goal has clear completion evidence (in which case call complete_eternal_goal), do not drift to unrelated optimizations just because this patrol was idle.',
+                statusStandby: 'Eternal · Standing by',
+                toolDescription: 'Mark the current Eternal goal as complete. Only call when there is concrete evidence the goal has been achieved. After calling, the goal is cleared, control mode auto-switches to manual, and Leader enters idle standby (user can set a new goal to continue). Do not call while parts of the goal are still unverified.',
+                toolEvidenceDescription: 'Concrete evidence of goal completion (passing tests, successful builds, committed code, or other verifiable facts)',
+                toolSummaryDescription: 'Completion summary including what was accomplished, key outputs, and follow-up recommendations',
+            },
+        },
+        workerTask: {
+            taskContextHeading: 'Task Context',
+            taskGoalLabel: 'Goal',
+            taskDescriptionLabel: 'Description',
+            workspaceLabel: 'Workspace',
+            writeScopeLabel: 'Write scope',
+            sessionLabel: 'Session',
+            budgetFast: '**Budget**: Quick task. Focus on direct evidence and minimal changes.',
+            budgetMedium: '**Budget**: Medium task. Understand first, then act, and verify after completion.',
+            budgetLarge: '**Budget**: Large task. Plan before execution; use Scratchpad to track high-complexity work.',
+            contextManifestTitle: '### Context Manifest (system-injected)',
+            leaderContextHeading: 'Leader Background Context (read before starting)',
+            leaderContextIntro: 'Provided by the Leader when assigning the task. It contains user intent, known facts, and key decisions:',
+            injectedSkillBodiesTitle: 'Injected Skill Bodies',
+            knowledgeGraphHeading: 'Knowledge Graph Protocol',
+            knowledgeGraphIntro: 'The shared knowledge graph is real-time public memory across Agents. First read the blackboard snapshot in Context Manifest; Facts/Intents include content, tags, evidence, and relationships.',
+            knowledgeGraphAction: 'Continue from existing facts; when you find new facts, new directions, or contradictions, append public memory with structured code blocks:',
+            skillPathRulesHeading: 'Skill Path Rules',
+            skillPriorityLine: ({ projectSkillsDir, globalSkillsDir }) => `Priority: project-level > plugin-contributed > user-level > built-in. Project directory: ${projectSkillsDir}/ | User directory: ${globalSkillsDir}/ | plugin/built-in directories follow the skills paths in Context Manifest`,
+            skillPathLine: 'Skill file paths must come from the real path field in the injected `<skill path="...">` block; read that path when the full skill is needed.',
+            autoInjectedLine: (names) => `Auto-injected: ${names.join(', ')}`,
+            skillBodiesLine: 'Skill bodies and plugin sources are included in Context Manifest; if truncated=true, read the full skill from the real path or referenced documents before producing critical office artifacts.',
+            deliverySopHeading: 'Delivery Integrity and Acceptance SOP',
+            deliveryScopeRule: 'If the task description covers only a subset but the user goal is clearly the whole project or complete feature, escalate to the Leader via send_message to align scope and continue planning against the complete goal',
+            deliveryCompletenessRule: 'Acceptance criteria include usability, integration path, error/empty/loading states, and traceable verification evidence; skeletons, placeholder pages, fake data, or disconnected features are stage artifacts only when explicitly required by the task contract',
+            deliveryPageEvidenceRule: 'Page-level tasks must provide both engineering evidence and real page evidence: `npm test`, typecheck, or successful build as engineering evidence, and browser acceptance as user-visible path evidence',
+            collaborationHeading: 'Collaboration and Guardrails',
+            collaborationRules: ({ taskId }) => [
+                'Previous Agent notes, artifacts, contracts, and file tree are in Context Manifest; read the manifest before starting, then use `read_work_notes` or manifest-referenced artifacts to fill evidence gaps',
+                `After meeting acceptance criteria, call \`attempt_completion\` to finish; every task must provide \`contract_compliance\` proof (surface/status/evidence/deviations). If there is no cross-stack contract, use \`task:${taskId}\` as the surface`,
+                'coding final summaries must include file paths and verification evidence; the system rejects complete when contract compliance proof is missing',
+                'After code changes, verify with compile (`tsc --noEmit`) or tests and confirm there are no errors',
+            ],
+            teamCommunicationHeading: 'Team Communication (if you belong to a team)',
+            teamCommunicationRules: [
+                'After starting the task, first call `team_inbox(unread_only=true, mark_read=true)`, directly handle existing P2P messages and team broadcasts, and incorporate teammate messages into your current judgment',
+                'Before starting, read the Context Manifest blackboard snapshot and prioritize `[contract]` / `[design_doc]` nodes; implement cross-stack work against existing contracts, and if missing or conflicting, notify the team/Leader first and submit a v+1 contract',
+                'When you need to confirm team members, call `team_manage(action="list_members")`; P2P targets must be `interactive=true` members: `team_message(target_type="member", target="member name", content=...)`. Use team broadcast for `not_dispatched` members, or use `send_message` to ask the Leader to dispatch them first',
+                'When you need teammate progress (who is doing what, which tasks are ready/blocked), call `team_manage(action="task_board")`, then proactively align write scopes, interface contracts, or integration windows',
+                'When you need to notify the whole team, call `team_message(target_type="team", target="team name", content=...)`; broadcasts are for interface contracts, global conventions, cross-module risks, integration windows, and other team-wide matters; always use `target_type + target` fields',
+                'If your output affects others, proactively notify them: new or changed API/data structures/file paths/component props/env vars/run commands/acceptance criteria require P2P or broadcast',
+                'When you need an explicit reply, use a complete call `team_message(target_type="member", target="<receiver>", content="...", type="request", request_id="<surface>@v<N>")`; after receiving and handling a request, you personally reply to the original sender with `team_message(target_type="member", target="<original sender>", content="...", type="ack", request_id="<same>")`. The member who handles the request sends the ack. `team_inbox` `ack_status` tells you who you are waiting on and whom you owe an ack',
+                'When you receive `[协商请求]`, it means you and another member have a file/resource conflict: negotiate directly first (split write paths or serialize work), then reply ack with the instruction request_id after agreement, and include target_type/target/content; escalate to the Leader only if negotiation fails',
+                'When cross-stack interface/schema/props/env/command contracts are added or changed, final output must include a `graph_contract` code block with at least `surface`, `title`, and `content`; use a `graph_design_doc` code block for important design decisions',
+                'When blocked by dependencies, contact the relevant member directly first; escalate to the Leader when user/Leader decisions, cross-team coordination, or timeout waiting is needed',
+                'Before finishing, call `team_inbox` again to ensure there are no unhandled collaboration messages or open acks; during wrap-up, use `write_work_note` to summarize key collaboration messages sent/received and mention them in the final result summary',
+                'Strategic reports (task wrap-up, Leader decision needed) use `send_message(recipient="leader", ...)`; when the task is complete, reply directly with the final result summary, and the system will automatically forward it to the Leader as the completion report',
+            ],
+            startInstruction: 'Start execution. After completion, reply directly with the result summary; when blocked, contact the Leader with send_message.',
+        },
+        sharedFragments: {
+            capabilitySurfaceHeading: 'Capability Surface Protocol (Plugin / Skill / MCP)',
+            capabilitySurfaceRules: [
+                'Plugin is a local capability package and distribution unit; when enabled, it contributes skills, MCP servers, apps/assets, tools/hooks/scripts to the Lingxiao capability surface',
+                'Skill is execution knowledge, workflow, and domain constraints injected into Agent context; source priority is project-level > plugin-contributed > user-level > built-in, and the Leader binds skills through skill_names in define_agent_role/create_task',
+                'MCP is the tools/resources/prompts protocol for connecting external systems; at runtime, discover and access tools/resources through the unified `mcp(action="list_servers|list_tools|call_tool|list_resources|read_resource", ...)` entrypoint',
+                'When prompts, resource templates, or service capability snapshots are needed, use the same `mcp` tool actions: action="list_prompts|get_prompt|list_resource_templates|capability_snapshot"',
+                'Plugin-contributed MCP servers are synced to settings.mcp.servers; all MCP servers are accessed through the same `mcp` tool entrypoint',
+                'Plugin tools/hooks/scripts belong to runtime callable capabilities only after being connected to ToolRegistry or an MCP server; otherwise they enter Context Manifest as marketplace, packaging, hook metadata, and local source clues',
+                'When the full Skill is needed, read the real path from the injected `<skill path="...">` block; before calling MCP, run list_servers/list_tools/list_resources, then call according to the real schema',
+                'Write files, commands, MCP resource URIs, URLs, screenshot/report paths, and verification results produced by capability calls into write_work_note, attempt_completion.evidence_refs, or lingxiao_completion.evidence_refs so Context Manifest can pass them to later Agents',
+            ],
+            externalCompletionHeading: 'Lingxiao Completion Report Protocol',
+            externalCompletionIntro: 'Write the final reply first as a natural-language summary for the Leader, then append a ```lingxiao_completion code block that Lingxiao parses and injects into later Agent context.',
+            externalCompletionSummaryExample: 'One sentence describing what was completed',
+            externalCompletionNoDeviationExample: 'none',
+            externalCompletionNextStepExample: 'Optional follow-up recommendation',
+            externalCompletionNotes: 'Use workspace-relative paths or absolute paths; contract_compliance is required proof of contract compliance and includes surface/status/evidence/deviations; evidence_refs should contain MCP resource URIs, URLs, screenshot/report paths, or external tool result IDs; blocked_by_discovery lists newly discovered unresolved dependencies, and needs_leader_coordination=true asks the Leader to coordinate; verification must record real commands and results; cross-stack interface/schema/props/env/command contract changes must also output a graph_contract code block.',
+            externalCompletionBrowserEvidence: 'Frontend/fullstack completion reports include real browser acceptance evidence; build, lint, or unit tests are engineering evidence, while page-level acceptance must list real page evidence separately. If the browser cannot start, write blocked/skipped and the error reason; mark visual or interaction results passed only when real page evidence is complete.',
+        },
+        tools: {
+            registry: {
+                schemaValidationFailed: 'schema validation failed',
+                deferredToolDescription: (name) => `Deferred tool: ${name}`,
+                argumentValidationMessage: (formatted) => `Argument validation failed: ${formatted}`,
+                argumentValidationFix: 'Fix arguments according to the tool schema; pass only declared fields and choose one target for mutually exclusive options. Omit unknown optional fields instead of passing empty strings.',
+                toolNotFoundMessage: (name) => `Tool not found: ${name}`,
+                toolNotFoundCause: (name) => `No tool is registered with the name ${name}.`,
+                toolNotFoundFix: (candidates) => candidates.length > 0 ? `Use one of these candidate tools instead: ${candidates.join(', ')}` : 'Check the tool name, or load/register the tool first.',
+                argumentParseFix: 'Tool arguments must be passed directly as a JSON object/array value; do not wrap arguments in a string.',
+                scopeForbiddenMessage: (name) => `Tool is not available in this context: ${name}`,
+                scopeForbiddenCause: (name) => `Tool ${name} can only be called from Leader context.`,
+                scopeForbiddenFix: 'Do not call Leader meta-tools directly from a Worker; report to the Leader for orchestration or use worker tools allowed by the task.',
+                validationFailedMessage: 'Argument validation failed.',
+                validationFailedFix: 'Fix arguments according to the tool schema.',
+                fileMustReadFirstMessage: (path) => `File must be read first: use file_read to read "${path}" and confirm its contents before editing.`,
+                fileMustReadFirstCause: 'To avoid overwriting unknown user changes, existing files must be read before editing or overwriting.',
+                fileMustReadFirstFix: 'Call file_read for this path first, then retry the edit.',
+                officeModeRequiredMessage: (name) => `Office tool requires Office mode: ${name}`,
+                officeModeRequiredCause: 'Office mode is not enabled for this session, so the office tool will not run.',
+                officeModeRequiredFix: 'Enable Office mode for this session with /office on or the frontend Office toggle, then retry.',
+                permissionRequiredFix: 'Wait for user authorization, or choose an alternative tool/approach that does not require this permission.',
+            },
+            structuredPatch: {
+                reuseOriginalReplace: '<reuse the original replace text here>',
+                reuseOriginalContent: '<reuse the original content text here>',
+                wholeLineRangeNote: 'Use only if replacing the whole listed line range is intended.',
+                exactOccurrenceNote: 'Use only if this exact occurrence is intended.',
+                finalOccurrenceNote: 'Use only if the final occurrence is intended.',
+                insertExactOccurrenceNote: 'Use only if inserting after this exact occurrence is intended.',
+                insertFinalOccurrenceNote: 'Use only if inserting after the final occurrence is intended.',
+                insertWholeRangeNote: 'Use only if inserting after the whole matched line/range is intended.',
+                retryRuleSingleShape: 'Retry with only {path,hunks}; each hunk must keep exactly one allowed shape. Do not copy received_fields back into retry args.',
+                ambiguousSearchFix: 'Prefer retry_args.first_occurrence or retry_args.last_occurrence to specify the target; use retry_args.replace_all only when you really want every match. If you only meant to append long text, use retry_args.append_eof directly.',
+            },
+            astQuery: {
+                actionDescription: 'AST-backed query action.',
+                symbolDescription: 'Symbol name for definitions, references, call_graph, or implementors.',
+                fileDescription: 'Optional project-relative file filter, for example src/index.ts.',
+                namePatternDescription: 'Regular expression applied to AST declaration names for action=pattern.',
+                kindsDescription: 'Declaration kinds for action=pattern.',
+                maxDepthDescription: 'Maximum call graph depth for action=call_graph.',
+                limitDescription: 'Maximum result count.',
+                symbolRequiredMessage: (action) => `ast_query action "${action}" requires a non-empty symbol.`,
+                symbolRequiredFix: (action) => `Retry with {"action":"${action}","symbol":"TargetName"}.`,
+            },
+        },
+        judges: {
+            nextSpeaker: {
+                toolDescription: 'Return whether an eternal-mode assistant response should continue immediately after a stop finish reason.',
+                system: [
+                    'You decide whether an autonomous leader in eternal mode should immediately continue after the model returned a stop/end_turn finish reason.',
+                    'Use semantic judgment, not local text-pattern rules.',
+                    'Continue only when the visible assistant output is substantively incomplete, has clearly started a structure it has not finished, or explicitly indicates it is mid-thought and must continue in the same turn.',
+                    'Do not continue merely because there may be future tasks; open work is handled separately by the runtime.',
+                    'Reply by calling submit_next_speaker_verdict exactly once.',
+                ].join(' '),
+            },
+            eternalPatrol: {
+                toolDescription: 'Decide whether the eternal-mode patrol should run an LLM patrol now, skip silently, or yield control to the user.',
+                actionDescription: 'patrol: project state has new signal worth investigating; skip: noise/already-handled change, do not spend LLM tokens; yield_user: nothing meaningful left to do, hand control back to the human.',
+                reasonDescription: 'Short justification (one sentence).',
+                system: [
+                    'You are the patrol judge for an Eternal Mode autonomous orchestrator.',
+                    'Decide if the leader should spend tokens running another patrol round, silently skip, or yield control back to the user.',
+                    'When eternal_goal is present, this is Goal Mode: protect that objective as the highest priority until there is concrete evidence it is complete or impossible without human-only input.',
+                    'Prefer patrol when project fingerprint moved AND there is a concrete failure, new task, new blackboard contract, or unanswered teammate message.',
+                    'With an active eternal_goal, also prefer patrol when the goal still needs planning, task creation, dispatch, implementation, verification, or final delivery, even if the project fingerprint did not move.',
+                    'Prefer skip when fingerprint changed but the diff looks like the leader\'s own previous output / heartbeat / cosmetic noise.',
+                    'Prefer yield_user when there is no eternal_goal AND fingerprint did NOT change AND last_patrol_outcome != productive AND has_open_work=false. Repeating generic idle patrols is wasted spend.',
+                    'Do not yield_user for an active eternal_goal merely because a round was idle or the fingerprint is unchanged.',
+                    'Reply only by calling submit_eternal_verdict exactly once.',
+                ].join(' '),
+            },
+            workerCompletion: {
+                toolDescription: 'Return whether the worker final text is a valid task completion.',
+                reasonDescription: 'Short stable reason for the verdict.',
+                feedbackDescription: 'Concrete feedback for the worker if completion is rejected; empty string if accepted.',
+                system: [
+                    'You are a completion judge for autonomous software-engineering workers.',
+                    'Use semantic judgment, not keyword matching.',
+                    'Accept when the worker has produced meaningful output relevant to the task — even if the output is a partial delivery, a summary, or a report that doesn\'t cover every requested detail.',
+                    'Only reject when the text is clearly still planning future work without having done anything, or hides an unresolved execution failure.',
+                    'Be lenient: if the worker has clearly made progress and produced substantive content, accept even if it\'s not perfect or complete in every aspect.',
+                    'Reply only by calling submit_completion_verdict exactly once.',
+                ].join(' '),
+            },
+            scratchpadReview: {
+                toolDescription: 'Return whether the scratchpad still contains meaningful unfinished follow-up work.',
+                system: [
+                    'You review session scratchpad notes for unfinished tail work.',
+                    'Use semantic judgment, not TODO keyword matching.',
+                    'Only mark follow-ups when the notes still imply real pending validation, delivery, cleanup, or blocked work.',
+                    'If all notes are complete, return has_follow_ups=false.',
+                    'Reply only by calling submit_scratchpad_followup_review exactly once.',
+                ].join(' '),
+            },
+            workflowCondition: {
+                taskInstructions: [
+                    'Evaluate the workflow condition using the input below.',
+                    'Return only one of these exact JSON values: true or false.',
+                    'Do not call tools. Do not explain.',
+                ],
+            },
+            taskClassification: {
+                toolDescription: 'Classify the task prompt mode that best matches the worker assignment.',
+                system: [
+                    'You classify worker task prompt mode for a software orchestration runtime.',
+                    'Use the assignment semantics, task context, and graph digest together.',
+                    'Choose bootstrap for initial framing or goal discovery work.',
+                    'Choose reason for situation analysis, gap synthesis, or planning the next exploration direction.',
+                    'Choose explore for executing a specific intent, investigation, validation, or implementation path.',
+                    'Choose generic for ordinary coding, research, review, QA, or delivery work that does not need a blackboard-specific prompt mode.',
+                    'Reply by calling submit_task_classification exactly once.',
+                ].join(' '),
+            },
+            intentClassification: {
+                toolDescription: 'Classify the intent of the user message.',
+                system: [
+                    'The Leader records user-message intent inside the same main LLM call; do not start an extra classifier LLM call.',
+                    'Judge the true intent from the full sentence semantics; do not use keyword matching, regex matching, or local token hits.',
+                    'record_capability_intent records a capability envelope, not a single label; primaryIntent is only a summary, while grants/denies/requiredGates/constraints drive gates.',
+                    'Fill primaryIntent, scope, phase, grants, denies, requiredGates, and constraints from the full-message semantics; do not use keyword matching.',
+                    'grants/denies may only use five coarse capabilities: read/write/shell/task/dispatch. read=file/search/analysis/planning; write=workspace file writes; shell=commands/git/npm/test/deploy/python/terminal; task=task graph; dispatch=worker dispatch.',
+                    'For read-only, explanation, or proposal requests, default to scope=read_only, grant only read, and deny write/shell/task/dispatch.',
+                    'For implementation or fix requests, grant read/write as authorized. If the user says no commands/git/deploy/npm/test, deny shell. If the user says no workers, deny dispatch.',
+                    'For complex/full-project work, use implement + project/workspace scope + design/prepare phase + read/write/task/dispatch grants, and include blueprint_coverage/verify_after_change gates as needed.',
+                    "For each new user turn, if record_capability_intent is available, call it once to record the profile; if it is unavailable or its result says the turn is already recorded, never call it again and continue executing the user's request directly.",
+                ].join(' '),
+            },
+            toolSelection: {
+                toolDescription: 'Select the tool names that should be exposed to the next agent LLM call.',
+                selectedToolNamesDescription: 'Tool names in priority order for the current task and role.',
+                system: [
+                    'You select tools for an autonomous software-engineering agent.',
+                    'Use the current role, task, workspace scope, and recent conversation to choose the tools that support the next LLM call.',
+                    'Return tool names in priority order. Include enough tools for planning, execution, verification, coordination, and completion when those capabilities fit the task.',
+                    'The runtime will enforce the token budget after your selection.',
+                    'Reply by calling submit_tool_selection exactly once.',
+                ].join(' '),
+            },
+        },
+    },
+};
+export function normalizePromptLocale(raw) {
+    return normalizeLanguage(raw);
+}
+export function resolvePromptLocale(input) {
+    if (input?.sessionLanguage)
+        return normalizePromptLocale(input.sessionLanguage);
+    if (input?.currentLanguage)
+        return normalizePromptLocale(input.currentLanguage);
+    return getSessionLanguage() || getLanguage();
+}
+export function getPromptLocale() {
+    return resolvePromptLocale({
+        sessionLanguage: getSessionLanguage(),
+        currentLanguage: getLanguage(),
+    });
+}
+export function getPromptCatalog(locale = getPromptLocale()) {
+    return catalogByLocale[locale];
+}
+export function getPromptLanguageDirective(locale = getPromptLocale()) {
+    return `\n\n${getPromptCatalog(locale).llmLanguageDirective}`;
+}
+//# sourceMappingURL=catalog.js.map

package/dist/agents/prompts/i18n/catalog.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"catalog.js","sourceRoot":"","sources":["../../../../src/agents/prompts/i18n/catalog.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,WAAW,EACX,kBAAkB,EAClB,iBAAiB,GAElB,MAAM,kBAAkB,CAAC;AAiK1B,MAAM,eAAe,GAAwC;IAC3D,EAAE,EAAE;QACF,MAAM,EAAE,IAAI;QACZ,oBAAoB,EAAE,mIAAmI;QACzJ,MAAM,EAAE;YACN,sBAAsB,EAAE,gBAAgB;YACxC,oBAAoB,EAAE,2FAA2F;YACjH,WAAW,EAAE;gBACX,gBAAgB,EAAE,gCAAgC;gBAClD,eAAe,EAAE,+BAA+B;gBAChD,YAAY,EAAE,4BAA4B;gBAC1C,aAAa,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,iBAAiB,WAAW,EAAE;gBAC9D,aAAa,EAAE,MAAM;gBACrB,eAAe,EAAE,qBAAqB;gBACtC,mBAAmB,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,cAAc,WAAW,oDAAoD;gBACnH,kBAAkB,EAAE,4EAA4E;gBAChG,6BAA6B,EAAE,yEAAyE;gBACxG,4BAA4B,EAAE,6DAA6D;gBAC3F,aAAa,EAAE,eAAe;gBAC9B,eAAe,EAAE,oHAAoH;gBACrI,uBAAuB,EAAE,iCAAiC;gBAC1D,sBAAsB,EAAE,0BAA0B;aACnD;SACF;QACD,UAAU,EAAE;YACV,kBAAkB,EAAE,OAAO;YAC3B,aAAa,EAAE,IAAI;YACnB,oBAAoB,EAAE,IAAI;YAC1B,cAAc,EAAE,KAAK;YACrB,eAAe,EAAE,MAAM;YACvB,YAAY,EAAE,IAAI;YAClB,UAAU,EAAE,2BAA2B;YACvC,YAAY,EAAE,4BAA4B;YAC1C,WAAW,EAAE,2CAA2C;YACxD,oBAAoB,EAAE,8BAA8B;YACpD,oBAAoB,EAAE,qBAAqB;YAC3C,kBAAkB,EAAE,qCAAqC;YACzD,wBAAwB,EAAE,uBAAuB;YACjD,qBAAqB,EAAE,QAAQ;YAC/B,mBAAmB,EAAE,oGAAoG;YACzH,oBAAoB,EAAE,2CAA2C;YACjE,qBAAqB,EAAE,YAAY;YACnC,iBAAiB,EAAE,CAAC,EAAE,gBAAgB,EAAE,eAAe,EAAE,EAAE,EAAE,CAAC,qCAAqC,gBAAgB,cAAc,eAAe,6CAA6C;YAC7L,aAAa,EAAE,kEAAkE;YACjF,gBAAgB,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;YACxD,eAAe,EAAE,mFAAmF;YACpG,kBAAkB,EAAE,cAAc;YAClC,iBAAiB,EAAE,2EAA2E;YAC9F,wBAAwB,EAAE,sEAAsE;YAChG,wBAAwB,EAAE,iEAAiE;YAC3F,oBAAoB,EAAE,OAAO;YAC7B,kBAAkB,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;gBAClC,4GAA4G;gBAC5G,6IAA6I,MAAM,IAAI;gBACvJ,iDAAiD;gBACjD,uCAAuC;aACxC;YACD,wBAAwB,EAAE,qBAAqB;YAC/C,sBAAsB,EAAE;gBACtB,6FAA6F;gBAC7F,2HAA2H;gBAC3H,sNAAsN;gBACtN,yFAAyF;gBACzF,2IAA2I;gBAC3I,6EAA6E;gBAC7E,+VAA+V;gBAC/V,iIAAiI;gBACjI,oIAAoI;gBACpI,2DAA2D;gBAC3D,gGAAgG;gBAChG,2GAA2G;aAC5G;YACD,gBAAgB,EAAE,+CAA+C;SAClE;QACD,eAAe,EAAE;YACf,wBAAwB,EAAE,6BAA6B;YACvD,sBAAsB,EAAE;gBACtB,qFAAqF;gBACrF,uHAAuH;gBACvH,2JAA2J;gBAC3J,oIAAoI;gBACpI,6EAA6E;gBAC7E,gJAAgJ;gBAChJ,8GAA8G;gBAC9G,wKAAwK;aACzK;YACD,yBAAyB,EAAE,UAAU;YACrC,uBAAuB,EAAE,yFAAyF;YAClH,gCAAgC,EAAE,YAAY;YAC9C,oCAAoC,EAAE,GAAG;YACzC,iCAAiC,EAAE,SAAS;YAC5C,uBAAuB,EAAE,4RAA4R;YACrT,iCAAiC,EAAE,yHAAyH;SAC7J;QACD,KAAK,EAAE;YACL,QAAQ,EAAE;gBACR,sBAAsB,EAAE,aAAa;gBACrC,uBAAuB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,IAAI,EAAE;gBACnD,yBAAyB,EAAE,CAAC,SAAS,EAAE,EAAE,CAAC,UAAU,SAAS,EAAE;gBAC/D,qBAAqB,EAAE,kEAAkE;gBACzF,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,SAAS,IAAI,EAAE;gBAC9C,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,IAAI,OAAO;gBAClD,eAAe,EAAE,CAAC,UAAU,EAAE,EAAE,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,mBAAmB;gBACnH,gBAAgB,EAAE,2CAA2C;gBAC7D,qBAAqB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,IAAI,EAAE;gBACrD,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,IAAI,oBAAoB;gBAC7D,iBAAiB,EAAE,gEAAgE;gBACnF,uBAAuB,EAAE,SAAS;gBAClC,mBAAmB,EAAE,kBAAkB;gBACvC,wBAAwB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,iCAAiC,IAAI,iBAAiB;gBAC1F,sBAAsB,EAAE,+BAA+B;gBACvD,oBAAoB,EAAE,iCAAiC;gBACvD,yBAAyB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,oBAAoB,IAAI,EAAE;gBAC/D,uBAAuB,EAAE,6BAA6B;gBACtD,qBAAqB,EAAE,qDAAqD;gBAC5E,qBAAqB,EAAE,0BAA0B;aAClD;YACD,eAAe,EAAE;gBACf,oBAAoB,EAAE,sBAAsB;gBAC5C,oBAAoB,EAAE,sBAAsB;gBAC5C,kBAAkB,EAAE,qBAAqB;gBACzC,mBAAmB,EAAE,oBAAoB;gBACzC,mBAAmB,EAAE,mBAAmB;gBACxC,yBAAyB,EAAE,uBAAuB;gBAClD,yBAAyB,EAAE,sBAAsB;gBACjD,oBAAoB,EAAE,wBAAwB;gBAC9C,oBAAoB,EAAE,8EAA8E;gBACpG,kBAAkB,EAAE,2IAA2I;aAChK;YACD,QAAQ,EAAE;gBACR,iBAAiB,EAAE,eAAe;gBAClC,iBAAiB,EAAE,0DAA0D;gBAC7E,eAAe,EAAE,8BAA8B;gBAC/C,sBAAsB,EAAE,qCAAqC;gBAC7D,gBAAgB,EAAE,yBAAyB;gBAC3C,mBAAmB,EAAE,6BAA6B;gBAClD,gBAAgB,EAAE,SAAS;gBAC3B,qBAAqB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,qBAAqB,MAAM,gBAAgB;gBAC9E,iBAAiB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,iBAAiB,MAAM,8BAA8B;aACrF;SACF;QACD,MAAM,EAAE;YACN,WAAW,EAAE;gBACX,eAAe,EAAE,qEAAqE;gBACtF,MAAM,EAAE;oBACN,6EAA6E;oBAC7E,sBAAsB;oBACtB,iEAAiE;oBACjE,4CAA4C;oBAC5C,yCAAyC;iBAC1C,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,aAAa,EAAE;gBACb,eAAe,EAAE,2DAA2D;gBAC5E,iBAAiB,EAAE,uFAAuF;gBAC1G,iBAAiB,EAAE,YAAY;gBAC/B,MAAM,EAAE;oBACN,sCAAsC;oBACtC,sDAAsD;oBACtD,6EAA6E;oBAC7E,8FAA8F;oBAC9F,0FAA0F;oBAC1F,wEAAwE;oBACxE,kIAAkI;oBAClI,sEAAsE;oBACtE,oCAAoC;iBACrC,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,gBAAgB,EAAE;gBAChB,eAAe,EAAE,kCAAkC;gBACnD,iBAAiB,EAAE,0BAA0B;gBAC7C,mBAAmB,EAAE,mCAAmC;gBACxD,MAAM,EAAE;oBACN,qCAAqC;oBACrC,kBAAkB;oBAClB,0DAA0D;oBAC1D,2CAA2C;oBAC3C,kDAAkD;oBAClD,uCAAuC;iBACxC,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,gBAAgB,EAAE;gBAChB,eAAe,EAAE,0CAA0C;gBAC3D,MAAM,EAAE;oBACN,2CAA2C;oBAC3C,uBAAuB;oBACvB,8CAA8C;oBAC9C,0CAA0C;oBAC1C,+CAA+C;iBAChD,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,iBAAiB,EAAE;gBACjB,gBAAgB,EAAE;oBAChB,oCAAoC;oBACpC,iCAAiC;oBACjC,cAAc;iBACf;aACF;YACD,kBAAkB,EAAE;gBAClB,eAAe,EAAE,iDAAiD;gBAClE,MAAM,EAAE;oBACN,wCAAwC;oBACxC,oDAAoD;oBACpD,0BAA0B;oBAC1B,+BAA+B;oBAC/B,mCAAmC;oBACnC,uEAAuE;oBACvE,wCAAwC;iBACzC,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,oBAAoB,EAAE;gBACpB,eAAe,EAAE,eAAe;gBAChC,MAAM,EAAE;oBACN,iDAAiD;oBACjD,sCAAsC;oBACtC,gIAAgI;oBAChI,sFAAsF;oBACtF,sKAAsK;oBACtK,gFAAgF;oBAChF,6GAA6G;oBAC7G,gKAAgK;oBAChK,uGAAuG;iBACxG,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,aAAa,EAAE;gBACb,eAAe,EAAE,wCAAwC;gBACzD,4BAA4B,EAAE,uBAAuB;gBACrD,MAAM,EAAE;oBACN,sBAAsB;oBACtB,4EAA4E;oBAC5E,+CAA+C;oBAC/C,mCAAmC;oBACnC,mCAAmC;iBACpC,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;SACF;KACF;IACD,EAAE,EAAE;QACF,MAAM,EAAE,IAAI;QACZ,oBAAoB,EAAE,yMAAyM;QAC/N,MAAM,EAAE;YACN,sBAAsB,EAAE,4BAA4B;YACpD,oBAAoB,EAAE,iMAAiM;YACvN,WAAW,EAAE;gBACX,gBAAgB,EAAE,kEAAkE;gBACpF,eAAe,EAAE,iEAAiE;gBAClF,YAAY,EAAE,yDAAyD;gBACvE,aAAa,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,2BAA2B,WAAW,EAAE;gBACxE,aAAa,EAAE,UAAU;gBACzB,eAAe,EAAE,uCAAuC;gBACxD,mBAAmB,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,iBAAiB,WAAW,yGAAyG;gBAC3K,kBAAkB,EAAE,oJAAoJ;gBACxK,6BAA6B,EAAE,sMAAsM;gBACrO,4BAA4B,EAAE,sKAAsK;gBACpM,aAAa,EAAE,uBAAuB;gBACtC,eAAe,EAAE,4TAA4T;gBAC7U,uBAAuB,EAAE,oHAAoH;gBAC7I,sBAAsB,EAAE,gGAAgG;aACzH;SACF;QACD,UAAU,EAAE;YACV,kBAAkB,EAAE,cAAc;YAClC,aAAa,EAAE,MAAM;YACrB,oBAAoB,EAAE,aAAa;YACnC,cAAc,EAAE,WAAW;YAC3B,eAAe,EAAE,aAAa;YAC9B,YAAY,EAAE,SAAS;YACvB,UAAU,EAAE,uEAAuE;YACnF,YAAY,EAAE,mFAAmF;YACjG,WAAW,EAAE,8FAA8F;YAC3G,oBAAoB,EAAE,wCAAwC;YAC9D,oBAAoB,EAAE,kDAAkD;YACxE,kBAAkB,EAAE,0GAA0G;YAC9H,wBAAwB,EAAE,uBAAuB;YACjD,qBAAqB,EAAE,0BAA0B;YACjD,mBAAmB,EAAE,gMAAgM;YACrN,oBAAoB,EAAE,6IAA6I;YACnK,qBAAqB,EAAE,kBAAkB;YACzC,iBAAiB,EAAE,CAAC,EAAE,gBAAgB,EAAE,eAAe,EAAE,EAAE,EAAE,CAAC,4FAA4F,gBAAgB,uBAAuB,eAAe,6EAA6E;YAC7R,aAAa,EAAE,+IAA+I;YAC9J,gBAAgB,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,kBAAkB,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;YACjE,eAAe,EAAE,iMAAiM;YAClN,kBAAkB,EAAE,uCAAuC;YAC3D,iBAAiB,EAAE,yNAAyN;YAC5O,wBAAwB,EAAE,oQAAoQ;YAC9R,wBAAwB,EAAE,0MAA0M;YACpO,oBAAoB,EAAE,8BAA8B;YACpD,kBAAkB,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;gBAClC,yMAAyM;gBACzM,yNAAyN,MAAM,mBAAmB;gBAClP,iJAAiJ;gBACjJ,mGAAmG;aACpG;YACD,wBAAwB,EAAE,8CAA8C;YACxE,sBAAsB,EAAE;gBACtB,6MAA6M;gBAC7M,iQAAiQ;gBACjQ,4TAA4T;gBAC5T,+MAA+M;gBAC/M,8RAA8R;gBAC9R,0LAA0L;gBAC1L,6fAA6f;gBAC7f,oTAAoT;gBACpT,iQAAiQ;gBACjQ,oLAAoL;gBACpL,0PAA0P;gBAC1P,oQAAoQ;aACrQ;YACD,gBAAgB,EAAE,gIAAgI;SACnJ;QACD,eAAe,EAAE;YACf,wBAAwB,EAAE,oDAAoD;YAC9E,sBAAsB,EAAE;gBACtB,mLAAmL;gBACnL,6PAA6P;gBAC7P,iPAAiP;gBACjP,6LAA6L;gBAC7L,wIAAwI;gBACxI,wOAAwO;gBACxO,wMAAwM;gBACxM,+QAA+Q;aAChR;YACD,yBAAyB,EAAE,qCAAqC;YAChE,uBAAuB,EAAE,sLAAsL;YAC/M,gCAAgC,EAAE,4CAA4C;YAC9E,oCAAoC,EAAE,MAAM;YAC5C,iCAAiC,EAAE,mCAAmC;YACtE,uBAAuB,EAAE,ojBAAojB;YAC7kB,iCAAiC,EAAE,wWAAwW;SAC5Y;QACD,KAAK,EAAE;YACL,QAAQ,EAAE;gBACR,sBAAsB,EAAE,0BAA0B;gBAClD,uBAAuB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,kBAAkB,IAAI,EAAE;gBAC3D,yBAAyB,EAAE,CAAC,SAAS,EAAE,EAAE,CAAC,+BAA+B,SAAS,EAAE;gBACpF,qBAAqB,EAAE,4LAA4L;gBACnN,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,mBAAmB,IAAI,EAAE;gBACxD,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,uCAAuC,IAAI,GAAG;gBAC3E,eAAe,EAAE,CAAC,UAAU,EAAE,EAAE,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,6CAA6C,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,uDAAuD;gBACvL,gBAAgB,EAAE,yGAAyG;gBAC3H,qBAAqB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,0CAA0C,IAAI,EAAE;gBACjF,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,IAAI,0CAA0C;gBACrF,iBAAiB,EAAE,uIAAuI;gBAC1J,uBAAuB,EAAE,6BAA6B;gBACtD,mBAAmB,EAAE,6CAA6C;gBAClE,wBAAwB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,mDAAmD,IAAI,4CAA4C;gBACvI,sBAAsB,EAAE,uGAAuG;gBAC/H,oBAAoB,EAAE,0DAA0D;gBAChF,yBAAyB,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,qCAAqC,IAAI,EAAE;gBAChF,uBAAuB,EAAE,+EAA+E;gBACxG,qBAAqB,EAAE,gGAAgG;gBACvH,qBAAqB,EAAE,4GAA4G;aACpI;YACD,eAAe,EAAE;gBACf,oBAAoB,EAAE,wCAAwC;gBAC9D,oBAAoB,EAAE,wCAAwC;gBAC9D,kBAAkB,EAAE,gEAAgE;gBACpF,mBAAmB,EAAE,gDAAgD;gBACrE,mBAAmB,EAAE,+CAA+C;gBACpE,yBAAyB,EAAE,gEAAgE;gBAC3F,yBAAyB,EAAE,+DAA+D;gBAC1F,oBAAoB,EAAE,uEAAuE;gBAC7F,oBAAoB,EAAE,gIAAgI;gBACtJ,kBAAkB,EAAE,sOAAsO;aAC3P;YACD,QAAQ,EAAE;gBACR,iBAAiB,EAAE,0BAA0B;gBAC7C,iBAAiB,EAAE,uEAAuE;gBAC1F,eAAe,EAAE,kEAAkE;gBACnF,sBAAsB,EAAE,yEAAyE;gBACjG,gBAAgB,EAAE,uCAAuC;gBACzD,mBAAmB,EAAE,iDAAiD;gBACtE,gBAAgB,EAAE,uBAAuB;gBACzC,qBAAqB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,qBAAqB,MAAM,gCAAgC;gBAC9F,iBAAiB,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,yBAAyB,MAAM,2BAA2B;aAC1F;SACF;QACD,MAAM,EAAE;YACN,WAAW,EAAE;gBACX,eAAe,EAAE,2GAA2G;gBAC5H,MAAM,EAAE;oBACN,6IAA6I;oBAC7I,sDAAsD;oBACtD,iNAAiN;oBACjN,2GAA2G;oBAC3G,4DAA4D;iBAC7D,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,aAAa,EAAE;gBACb,eAAe,EAAE,mHAAmH;gBACpI,iBAAiB,EAAE,mMAAmM;gBACtN,iBAAiB,EAAE,qCAAqC;gBACxD,MAAM,EAAE;oBACN,uEAAuE;oBACvE,0HAA0H;oBAC1H,yLAAyL;oBACzL,kJAAkJ;oBAClJ,6MAA6M;oBAC7M,8HAA8H;oBAC9H,+LAA+L;oBAC/L,+GAA+G;oBAC/G,4DAA4D;iBAC7D,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,gBAAgB,EAAE;gBAChB,eAAe,EAAE,kEAAkE;gBACnF,iBAAiB,EAAE,sCAAsC;gBACzD,mBAAmB,EAAE,uFAAuF;gBAC5G,MAAM,EAAE;oBACN,yEAAyE;oBACzE,8CAA8C;oBAC9C,2LAA2L;oBAC3L,yIAAyI;oBACzI,qJAAqJ;oBACrJ,+DAA+D;iBAChE,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,gBAAgB,EAAE;gBAChB,eAAe,EAAE,oFAAoF;gBACrG,MAAM,EAAE;oBACN,+DAA+D;oBAC/D,mDAAmD;oBACnD,8GAA8G;oBAC9G,yDAAyD;oBACzD,uEAAuE;iBACxE,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,iBAAiB,EAAE;gBACjB,gBAAgB,EAAE;oBAChB,wDAAwD;oBACxD,4DAA4D;oBAC5D,oCAAoC;iBACrC;aACF;YACD,kBAAkB,EAAE;gBAClB,eAAe,EAAE,wEAAwE;gBACzF,MAAM,EAAE;oBACN,4EAA4E;oBAC5E,wEAAwE;oBACxE,8DAA8D;oBAC9D,kGAAkG;oBAClG,oGAAoG;oBACpG,kIAAkI;oBAClI,2DAA2D;iBAC5D,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,oBAAoB,EAAE;gBACpB,eAAe,EAAE,0CAA0C;gBAC3D,MAAM,EAAE;oBACN,kHAAkH;oBAClH,2HAA2H;oBAC3H,yKAAyK;oBACzK,gJAAgJ;oBAChJ,sPAAsP;oBACtP,oIAAoI;oBACpI,+KAA+K;oBAC/K,sMAAsM;oBACtM,qPAAqP;iBACtP,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;YACD,aAAa,EAAE;gBACb,eAAe,EAAE,0EAA0E;gBAC3F,4BAA4B,EAAE,6DAA6D;gBAC3F,MAAM,EAAE;oBACN,gEAAgE;oBAChE,0HAA0H;oBAC1H,qKAAqK;oBACrK,iEAAiE;oBACjE,sDAAsD;iBACvD,CAAC,IAAI,CAAC,GAAG,CAAC;aACZ;SACF;KACF;CACF,CAAC;AAEF,MAAM,UAAU,qBAAqB,CAAC,GAAW;IAC/C,OAAO,iBAAiB,CAAC,GAAG,CAAC,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAGnC;IACC,IAAI,KAAK,EAAE,eAAe;QAAE,OAAO,qBAAqB,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;IAChF,IAAI,KAAK,EAAE,eAAe;QAAE,OAAO,qBAAqB,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;IAChF,OAAO,kBAAkB,EAAE,IAAI,WAAW,EAAE,CAAC;AAC/C,CAAC;AAED,MAAM,UAAU,eAAe;IAC7B,OAAO,mBAAmB,CAAC;QACzB,eAAe,EAAE,kBAAkB,EAAE;QACrC,eAAe,EAAE,WAAW,EAAE;KAC/B,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,SAAuB,eAAe,EAAE;IACvE,OAAO,eAAe,CAAC,MAAM,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,SAAuB,eAAe,EAAE;IACjF,OAAO,OAAO,gBAAgB,CAAC,MAAM,CAAC,CAAC,oBAAoB,EAAE,CAAC;AAChE,CAAC"}

package/dist/agents/prompts/i18n/leader_system_prompt.d.ts

@@ -0,0 +1,6 @@
+import type { PromptLocale } from './catalog.js';
+export declare const ZH_LEADER_SYSTEM_PROMPT: string;
+export declare const EN_LEADER_SYSTEM_PROMPT: string;
+export type LeaderPromptProfile = 'solo' | 'team' | 'workflow';
+export declare function getLeaderSystemPromptTemplate(locale: PromptLocale, profile?: LeaderPromptProfile): string;
+//# sourceMappingURL=leader_system_prompt.d.ts.map

package/dist/agents/prompts/i18n/leader_system_prompt.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"leader_system_prompt.d.ts","sourceRoot":"","sources":["../../../../src/agents/prompts/i18n/leader_system_prompt.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAOjD,eAAO,MAAM,uBAAuB,QAsT5B,CAAC;AAET,eAAO,MAAM,uBAAuB,QA0S5B,CAAC;AAET,MAAM,MAAM,mBAAmB,GAAG,MAAM,GAAG,MAAM,GAAG,UAAU,CAAC;AAyT/D,wBAAgB,6BAA6B,CAAC,MAAM,EAAE,YAAY,EAAE,OAAO,GAAE,mBAA4B,GAAG,MAAM,CAEjH"}