DeliveryPilot Conduit
PM Requirement → Verified Delivery Evidence
未导入仓库
TEST MODE · Local writes only for sandbox · No push / no real GitHub PR · Official sandbox only
AI Delivery Cockpit
让 PM 用一句需求驱动 Conduit 真实代码交付

Prompt in. Verified patch out.

输入产品需求,DeliveryPilot 会完成澄清、PRD、模块定位、Patch、测试与 Evidence,全程可人工接管。

Advanced actions
Workspace details
Public URL
Skill Countloading
Benchmark ScoreRun benchmark
Public
Blocked
HTTPS
Last Smoke Test
Official L1 Demo Path
3-minute guided path
Official L1 Demo Path
not started
点击 Run Official L1 Demo,沿官方 L1 wordcount 链路进入 Delivery Flow 并查看 PR Evidence。
Demo path 正在等待 workspace summary。
Practice suite

Three L1 paths to prove speed, safety, and reuse.

L1 · frontend-only Ready

Article reading stats

给文章详情页增加字数统计和预计阅读时间,不触碰后端与数据库。

Article detail No schema change Fast verifier
verify-official-l1-wordcount-patch
L1 · frontend-only Ready

Popular Tags Top 5

给 Popular Tags 前 5 个标签增加视觉标识,不改变排序和点击文本。

Popular tags No API change Click-safe badge
verify-official-l1-popular-tags-patch
L1 · frontend-only Ready

Profile About Me Tab

在个人主页增加 About Me Tab,展示 User.bio,并为空 bio 提供占位。

Profile tab No backend Skill hot-swap
verify-official-l1-profile-about-tab-patch
Advanced practice

L2 / L3 cases for cross-stack and reasoning quality.

L2 · Cross-stack Full-chain

Article Cover Image

Article.coverImage 从后端模型、API 输入输出、前端表单、列表卡片、详情页到测试层全部同步。

Model API Editor List Detail Tests
verify-official-l2-cover-image-patch
L3 · Ambiguous requirement Gate first

Draft Workflow Clarifier

先澄清可见性、作者权限、发布动作、公开列表过滤和 feed 边界,再允许进入下游 patch。

Visibility Owner Publish Lists Review Replay
verify-l3-draft-status-no-premature-patch
Open full delivery workspace Repository, Flow, Patch, Test, Evidence, Raw JSON
Advanced / Raw deployment evidence · Deployment Status · Deployment Evidence
Deployment evidence loads from the public smoke check. If this stays empty, click Refresh Status.
仓库导入
Start with the official Conduit sandbox
公网 Demo 默认使用服务器内置实验田,不需要输入服务器路径。
仅本机开发模式使用自定义路径;公网 DEMO_MODE 下此入口会被禁用,请使用上方 Official Conduit Sandbox。
文件树
0 files
点击 Use Official Conduit Sandbox 生成 File Tree、Repo Overview 和 Module Map。
仓库概览
Import the official sandbox to show repo summary, scripts, modules, and evidence-ready context.
package scripts
Package scripts appear after repository indexing.
模块地图
Module map appears after Official Conduit Sandbox import.
搜索
导入仓库后搜索 article / profile / comment / prisma,快速定位上下文。
交付流水线(Stage 2:Contract → Plan → Location → Evidence)
Current Delivery
Start with a PM requirement
先输入需求或运行 Official L1 Demo。这里会显示当前阶段、下一步动作和证据状态。
Stage not started
运行 Official L1 Demo 或创建需求后显示 Progress / Completed / Blocked / Evidence。
未创建会话
创建需求后,这里会按阶段显示 PRD、Plan、Context、Patch、Test 和 PR Evidence 的摘要。
澄清问题
创建需求后点击“生成澄清问题”,确认 PM 输入是否足够进入 PRD。
PRD 契约(可编辑)
方案拆解(JSON,可编辑)
模块定位(Impact Heatmap)
Target Files
确认方案后点击“模块定位”,生成 target files。
Test Files
模块定位后显示建议测试文件。
Do Not Touch
模块定位后显示 doNotTouch 边界。
人工边界调整
Evidence Ledger
运行 Official L1 Demo 或刷新当前需求证据账本后显示 workflow events。
Observability(P1)
Observability
AI calls not loaded
刷新后展示 AI 调用次数、fallback、latency、tokens 和 cost;原始日志保留下方。
not started View Logs
Skill Marketplace(Stage 5)
Skill Marketplace
Skills not loaded
刷新 Skill 列表后,系统会展示可复用交付技能和 dry-run 匹配结果。
not started View Skill Evidence
Skills
点击“刷新 Skill 列表”
Dry Run
Context Strategy Lab(Stage 5)
Context Strategy Lab
Strategy comparison not run
创建需求后比较 minimal / balanced / defensive,先看推荐策略,再查看原始 JSON。
not started View Context JSON
点击 “Compare Strategies” 生成 minimal / balanced / defensive 对比
Delivery Memory Loop(Stage 5)
Delivery Memory Loop
Memory suggestion not generated
写入或召回历史交付经验后,这里会显示 reuse decision、confidence 和可应用范围。
not started View Memory JSON
Memory Library
Suggestion
说明:建议包含 suggestedSkill / contextStrategy / selectedFiles / excludedFiles / doNotTouch / testStrategy / confidence / reuseDecision,并会写入 workflow events(memory_written / memory_suggested / memory_applied)。
Requirement Benchmark Suite(Stage 5)
Benchmark Cases
点击 Refresh Cases 查看 L1 / L2 / L3 标准需求集。
Scoreboard
Run All Quick to produce average score, by-level scores, and weak areas.
Selected Case Detail
Select a benchmark case to inspect expected capability profile and failures.
说明:Benchmark v1 目标是评估 Skill/Clarification/PRD/Plan/Context/Cross-stack Gate(coverImage)能力,不要求所有 case 都生成 patch。L3 case 必须体现“先澄清,不直接 patch”。
Demo Scenario Runner
Scenario List
点击 Refresh Scenarios 后选择 official L1、coverImage review 或 memory reuse 场景。
Run Detail
Run Selected 后展示 scenario run、状态和 evidence refs。
Step Timeline
运行场景后显示每一步:requirement、context、patch、test、review。
Evidence Links
Scenario evidence links appear after a run.
说明:Demo Scenario Runner 将 Skill、Context、Memory、Benchmark、Review、Replay 串成可一键运行的回归/演示场景,输出 ordered steps 与 evidenceRefs。
Human Review Workbench(P1)
Review Detail
Failure reasons, evidence, actions, decisions
Failure Reasons
选择队列项后展示失败原因。
Evidence Refs
选择队列项后展示 evidence refs。
Recommended Actions
选择队列项后展示建议动作。
Decision History
暂无人工决策
Decision(for selected item)
Replay Panel
Request Only / Dry Run / Execute Downstream / Rollback
Request Only 记录人工重放请求,不改下游产物。
Dry Run 生成 Artifact Diffs 预览,不写正式 gate。
Execute Downstream 需 approval code,执行下游重放。
Rollback 回滚已执行的 replay run。
Affected Stages
Request Replay 后显示受影响阶段。
Safety Checks
Dry Run 或 Execute 前显示安全检查。
Artifact Diffs
Dry Run 后显示 artifact diffs。
Replay History
执行 replay 后显示历史记录。
Controlled Replay JSON
Selected Item Detail JSON
说明:Review Workbench 以 Benchmark 失败/弱点队列为主,支持 approve/reject/request-replay/close;Controlled Replay 支持 request_only、dry_run、execute_downstream 和 rollback,execute 需要人工 approval code。
Evaluation Dashboard(P1)
Benchmark Scoreboard
Refresh Evaluation Dashboard to load benchmark average, cross-stack gate, replay, strict verification, and validation status.
Capability Metrics
Capability metrics load Skill Match, Context Strategy, Memory Reuse, and Review Queue signals.
AI Cost Summary
AI call count, estimated tokens, latency, and cost appear after refresh.
Capability Validation Suite
Real AI Capability Test
fallback
Failed Cases
Run Capability Suite to inspect failed cases. In public demo mode, real AI live calls remain CLI-only.
Real AI / Cost Status
AI_MODE and ARK_* are read only on the server. Secret values are never returned to the browser.
说明:Evaluation Dashboard 聚合 Benchmark Scoreboard、Skill Accuracy、Context Strategy Distribution、Memory Reuse Decisions、Review Queue Open Items、Cross-stack Gate、Replay success rate、Strict verification status、Validation pass rate 和 AI latency/usage/cost。
Governance Policy Engine(P1)
Governance Policy Engine
Policies not evaluated
刷新 policies 后可查看 scope、secret、cross-stack 和 human approval 门禁;Evaluate 后显示 pass/block 摘要。
not started View Policy JSON
Policies
Policy Input JSON
Decision Summary
说明:Policy Engine 将 scope guard、secret hygiene、cross-stack gate 和 human approval 统一为可审计 policy decisions,并保留 evidenceRefs。
Final Evidence Center(P1)
Coverage Matrix
Refresh Evidence Center to see Requirement / Clarification / Plan / Context / Patch / Test / PR / Review / Replay / Deployment coverage.
Verification Commands
Verification commands appear after evidence refresh.
Export Result
Export JSON + Markdown + HTML after the evidence center is refreshed.
说明:Final Evidence Center 汇总 official L1、coverImage cross-stack、Memory、Benchmark、Review、Controlled Replay、Observability 和 strict verification 证据,并支持离线 JSON/Markdown/HTML 导出。
Patch Theatre(Stage 3)
Patch Theatre
Patch not generated
先确认方案和模块边界,再生成 patch。原始 diff 会保留在下方。
not started View Diff JSON
Unified Diff
Spec-to-Diff Trace Matrix
生成 patch 后显示
DoNotTouch / Scope Guard
Test Arena(Stage 3)
Test Arena
Tests not run
Patch apply 后生成 Test Plan,再运行白名单测试。
not started View Logs JSON
QA Report(Stage 3)
PR & CI Manager
PR Evidence
PR package not created
测试通过后创建 PR Package;公网 Demo 不创建真实 GitHub PR。
not started View PR JSON
PR Record
CI Status
说明:package 模式生成 branch / commitSha / patch.diff / pr-summary.md / validation.log / rollback-plan.md;没有真实 GitHub 集成时不会伪造 PR URL,而是记录 GITHUB_PR_BLOCKED。
Preview Deploy(Stage 4)
Preview Deploy
Preview not deployed
先生成并确认 Deployment Plan,再执行 preview / smoke / release note。
not started View Deploy JSON
Deployment Plan
Deployment / Smoke
Native Conduit Preview Adapter
Native Preview
Native preview not checked
公网 Demo 默认阻断 native start;这里只展示真实依赖检查和 blocker reason。
not started View Native JSON
Native Preview Plan / Status
Dependency / Health Check
说明:contract preview 仍是快速稳定路径;native preview 只在真实 health check 通过后显示 URL,否则记录 NATIVE_PREVIEW_BLOCKED,不伪造成功。
Knowledge Flywheel(Stage 4)
Knowledge Flywheel
Knowledge not written
QA evidence ready 后写入知识,后续需求可复用 playbook。
not started View Knowledge JSON
Reuse Replay(Stage 4)
Reuse Replay
Reuse not evaluated
输入相似需求并 Find Similar,验证历史知识是否能指导新交付。
not started View Reuse JSON
Evidence Hub(Stage 4)
Evidence Hub
Evidence hub not refreshed
刷新后汇总 requirement、PRD、plan、module location、patch、test、deploy 和 knowledge 证据。
not started View Hub JSON