这份路线图按当前代码状态重写,目标是明确:
- 哪些能力已经落地
- 哪些能力已经进入 Alpha 可用
- 哪些能力仍然是下一阶段重点
ResearchClaw 现在更准确的定义是:
- 一个本地优先的 Research OS
- 一个带控制面、消息通道、Automation、Cron、Skills、MCP 的长期运行 Agent 平台
- 一个已经有 project / workflow / claim / evidence / experiment / artifact 主链路的科研工作流系统
一句话目标仍然不变:
用户在 Console 或 IM 中提出研究任务,系统持续推进,并把过程沉淀成可追踪、可恢复、可复用的科研状态。
- Gateway Lite 与控制面
app/gateway/runtime.pyapp/gateway/health.pyapp/gateway/schemas.py/api/control/*
- 多 Agent / channels / providers / skills / MCP
- multi-agent runner
- bindings / channel accounts
- provider fallback chains
SKILL.md+ Skills Hub- MCP client 管理
- Research service / runtime 已接入 gateway
research_serviceresearch_runtime- proactive cycle 与 health snapshot
- Research project 抽象
- project
- workflows
- notes
- experiments
- claims
- artifacts
- drafts
- Research workflow runtime
- stage / task / artifact / status / dependency
- workflow create / pause / resume / cancel / retry
- session / channel / automation / cron binding
- 持久化 store
- 结构化 stage worker
literature_searchpaper_readingnote_synthesishypothesis_queueexperiment_planexperiment_runresult_analysiswriting_tasksreview_and_followup
- Claim / evidence graph
- claim / evidence / source / artifact schema
- paper / note / experiment / citation / generated artifact 关联
- claim graph 查询 API
- Experiment tracking
- run schema
- parameters / inputs / metrics / outputs / notes
- baseline / ablation / comparison
- execution binding / event timeline / heartbeat / result ingest
- artifact contract / result bundle validation
- Structured research notes
- paper note
- idea note
- experiment note
- writing note
- decision log
- Proactive automation
- stale workflow reminder
- blocked workflow reminder
- remediation task follow-up
- paper watch reminder
- batch dispatch / execute / resume
- Console Research 页面
- project dashboard
- execution health
- recent blockers
- claim graph
- remediation detail / batch actions
- [~] 文献调研闭环
- 已有 literature search / paper reading / notes / claims
- 还缺高质量去重、打分、evidence matrix、related-work 产物
- [~] 证据链
- 已有 claim/evidence graph 与查询
- 还缺更严格的 claim-evidence validator 与冲突分析
- [~] 实验执行
- 已有 local / command / notebook / external / file-watch execution binding
- 还缺更强的外部 runner adapter、queue integration、notebook orchestration
- [~] 写作链
- 已有 writing_tasks / review_and_followup / draft artifact
- 还缺章节级写作器、reviewer mode、submission packaging
- [~] 长期陪伴
- 已有 project dashboard、blockers、remediation、proactive cycle
- 还缺更完整的 project-level timeline、reporting、复盘视图
- evidence matrix 与跨论文证据聚合
- 强约束的 claim-evidence validator
- reproducibility gate
- submission bundle / camera-ready packaging
- 更系统的 execution backend catalog
- 学科/场景化 research agent 编排模板
状态:已完成到可用阶段。
范围:
- gateway-lite 单体边界
- control plane
- channels
- providers / skills / MCP / automation / cron
状态:已完成到 Alpha 阶段。
范围:
- project / workflow / task / note / claim / evidence / experiment / artifact schema
- 持久化 store
- API 与 console 基本可视化
状态:已完成最小闭环。
范围:
- 结构化 stage worker
- workflow 持续推进
- claim/evidence/notes/experiments 回写
- blocker 与 remediation
状态:已完成最小主链,继续补强。
已完成:
- execution binding
- external result ingest
- local command/notebook launcher
- result bundle ingest
- contract validation
待补强:
- 更强 execution adapters
- runner preset / environment catalog
- reproducibility 和 replay
状态:已完成 Alpha 主链。
已完成:
- project dashboard
- execution health
- recent blockers
- per-task / per-workflow / per-project dispatch / execute / resume
待补强:
- project timeline
- richer review surfaces
- better reporting/export
状态:未完成。
目标:
- evidence matrix
- stronger validation
- reviewer mode
- submission bundle
- reproducibility checklist
优先补:
- evidence matrix
- claim-evidence conflict detection
- cross-paper synthesis ranking
优先补:
- queue / remote executor adapter
- notebook execution contract
- richer result bundle schema registry
优先补:
- section-level writing tasks
- reviewer / revision loop
- submission bundle 与 reproducibility gate
优先补:
- timeline / weekly summary
- blocker drill-down
- project report / export
当前阶段不优先做:
- 独立 gateway 进程
- 多节点控制面
- 通用 OpenAI facade gateway
- 只靠 prompt、没有状态层的“伪 workflow”
- 脱离科研闭环价值的表面 UI 美化