| name | project-roadmap |
| description | Analyze the current Pulse Coder project/repository and turn vague or concrete needs into prioritized requirements, milestones, tasks, validation plans, and execution handoff. Use when user asks “分析当前项目”, “后续需求推进”, “项目规划”, “下一步做什么”, “roadmap”, “需求拆解”, “排期”, or wants a project health check plus delivery plan. |
| version | 1.0.0 |
| author | Pulse Coder Team |
Project Roadmap Skill — 项目分析与需求推进
Goal
Use this skill to analyze the current project/repository and convert product or engineering needs into an actionable delivery plan.
The expected outcome is not just a summary. It must produce:
- Current-state project map.
- Evidence-backed diagnosis.
- Requirement clarification and decomposition.
- Prioritized roadmap.
- Milestone/task plan with validation gates.
- Clear next action for the agent or human team.
Trigger Examples
Use this skill when the user says things like:
- “帮我分析当前项目,并安排后续需求推进”
- “看一下这个项目现在该做什么”
- “给这个需求做拆解和排期”
- “帮我规划下一阶段 roadmap”
- “项目体检一下,列出优先级”
- “analyze this repo and propose next milestones”
Non-Negotiable Rules
-
Evidence first.
Project claims must come from files, commands, docs, tests, git history, or explicit user input. If not verified, mark as Assumption.
-
Protect user work.
Always inspect git status --short before proposing or making code changes. Never overwrite uncommitted user changes unless explicitly instructed.
-
Separate planning from implementation.
- Analysis and planning may proceed directly.
- Code changes, deletions, migrations, production operations, and dependency updates require explicit scope clarity. If the user only asked for planning, do not implement.
-
Ask only high-leverage questions.
If critical information is missing, ask at most 3 focused questions. Otherwise proceed with stated assumptions.
-
Make requirements testable.
Every proposed requirement must include acceptance criteria or a measurable success signal.
-
Prefer small safe increments.
Break work into reviewable PR-sized slices. Avoid giant “rewrite everything” plans unless the repo evidence makes it unavoidable.
-
Keep one active execution thread.
For multi-step work, use task tracking: list/create/update tasks, keep one main task in_progress, mark blockers clearly.
Current Project Profile: Pulse Coder
When operating in this repository, treat the project as a TypeScript pnpm monorepo.
Key Workspaces
| Path | Role |
|---|
packages/engine | Core agent engine: loop, tools, context, plugins, built-in skill/task/tool-search systems. |
packages/cli | Interactive terminal CLI built on pulse-coder-engine. |
packages/pulse-sandbox | Sandboxed JavaScript runtime and run_js tool adapter. |
packages/memory-plugin | Memory integration/service package. |
packages/plugin-kit | Runtime plugin toolkit, including worktree/vault/devtools helpers. |
packages/acp | ACP client/runner integration. |
packages/orchestrator | Multi-agent orchestration layer. |
packages/agent-teams | Multi-session collaborative agent coordination. |
packages/canvas-cli | Pulse Canvas CLI and bundled canvas skills. |
packages/langfuse-plugin | Observability plugin. |
apps/remote-server | HTTP/webhook runtime for Feishu/Discord and internal agent API. |
apps/teams-cli | CLI for agent teams. |
apps/canvas-workspace | Electron canvas workbench. |
Important Runtime Notes
- Source lives in each workspace’s
src/; build output goes to dist/.
- Root commands:
pnpm run build
pnpm test
pnpm run dev
pnpm start
- Useful targeted checks:
pnpm --filter pulse-coder-engine typecheck
pnpm --filter pulse-coder-engine test
pnpm --filter pulse-coder-cli test
pnpm --filter pulse-sandbox test
pnpm --filter pulse-coder-memory-plugin test
pnpm --filter @pulse-coder/remote-server build
apps/remote-server is the webhook/server runtime. Its key files include:
src/index.ts
src/server.ts
src/core/dispatcher.ts
src/core/agent-runner.ts
src/core/engine-singleton.ts
src/routes/internal.ts
- Skill files use
.pulse-coder/skills/<skill-name>/SKILL.md with YAML frontmatter.
Common Feature Areas
| Need type | Start investigation here |
|---|
| Agent loop behavior | packages/engine/src/core/loop.ts, packages/engine/src/context/, packages/engine/src/ai/ |
| Built-in tools | packages/engine/src/tools/ |
| Plugin loading/hooks | packages/engine/src/plugin/, packages/engine/src/built-in/ |
| Skills system | packages/engine/src/built-in/skills-plugin/, .pulse-coder/skills/ |
| Task tracking | packages/engine/src/built-in/task-tracking-plugin/ |
| Tool search/deferred tools | packages/engine/src/built-in/tool-search-plugin/, remote tool registration |
| CLI UX | packages/cli/src/ |
| Remote Discord/Feishu runtime | apps/remote-server/src/core/, apps/remote-server/src/platforms/, apps/remote-server/src/routes/ |
| Internal scheduled runs | apps/remote-server/src/routes/internal.ts, cron-related tools/routes |
| Memory | packages/memory-plugin/, apps/remote-server/src/core/memory-integration.ts |
| Worktree/vault binding | packages/plugin-kit/, apps/remote-server/src/core/worktree/ |
| Canvas workbench | apps/canvas-workspace/, docs/canvas-workspace-product/, docs/06-harness-engineering-roadmap.md |
Required Workflow
Phase 0 — Scope Alignment
Identify the user’s intent:
| User intent | Deliverable |
|---|
| “分析当前项目” | Project map + health diagnosis + risks + immediate next steps. |
| “需求推进” | Requirement decomposition + priority + milestone plan + task board. |
| “做某个功能” | PRD-lite + technical plan + implementation slices + validation plan. |
| “排期/roadmap” | Sequenced roadmap with dependencies, owners/roles, confidence, and risks. |
| “直接干” | Create execution tasks, inspect repo, implement only after scope is clear. |
If the user does not specify a target area, default to repo-wide analysis plus next 2–4 week plan.
Ask up to 3 questions only if needed:
- What is the target audience/user or business goal?
- What timeframe or release deadline matters?
- Are we planning only, or should we also implement the first slice?
If unanswered, continue with assumptions.
Phase 1 — Repository Discovery
Run or inspect, as appropriate:
pwd, git status --short, git branch --show-current.
- Workspace manifests:
package.json
pnpm-workspace.yaml
- each relevant workspace
package.json
- Project guidance:
README.md, README-CN.md
AGENTS.md, CLAUDE.md
docs/, architecture/
- Source topology:
- top-level
packages/*/src
- top-level
apps/*/src
- Existing tests and scripts:
*.test.ts, *.spec.ts
- package
scripts
- Recent work context when useful:
git diff --stat
git log --oneline -10
Do not read huge generated folders such as dist/, node_modules/, release outputs, session stores, or private memory data unless directly relevant.
Phase 2 — Project Map
Produce a concise system map:
- What the product is — one paragraph.
- Architecture layers — runtime, tools, plugins, clients/apps, storage/integration.
- Key modules and ownership boundaries.
- Primary data/control flows, for example:
- user input → platform/CLI → dispatcher/runner → engine loop → tools/plugins → response streaming
- skill loading → registry →
skill tool → detailed instructions
- Operational entrypoints and important commands.
- Known docs vs code mismatches if any.
Mark each claim as:
Fact — verified from repo evidence.
Inference — reasoned from evidence.
Assumption — not verified.
Phase 3 — Health Diagnosis
Assess the project across these dimensions:
| Dimension | Checkpoints |
|---|
| Product clarity | Who is the user? What job-to-be-done? What is the current wedge? |
| Architecture | Boundaries, dependency direction, plugin extensibility, runtime coupling. |
| Code quality | Type strictness, duplication, naming, local complexity, generated artifacts. |
| Testability | Existing tests, critical paths covered, missing integration/e2e tests. |
| Developer experience | Build scripts, targeted commands, docs, setup friction. |
| Runtime reliability | retries, abort, compaction, tool failure handling, state persistence. |
| Security | secrets handling, internal API protection, path/file operations, webhook verification. |
| Observability | logs, run IDs, latency traces, devtools, Langfuse/plugin hooks. |
| Delivery risk | blockers, unclear requirements, migrations, third-party APIs, platform quirks. |
Use severity:
P0 — blocks correctness/security/release.
P1 — important for near-term delivery.
P2 — improves maintainability or UX.
P3 — future optimization.
Phase 4 — Requirement Intake and Decomposition
For each need, create a PRD-lite block:
## Requirement: <name>
- Problem:
- Target user / scenario:
- Desired outcome:
- Non-goals:
- Constraints:
- Acceptance criteria:
1. ...
2. ...
- Success metrics:
- Dependencies:
- Risks:
- Open questions:
Then decompose into engineering slices:
### Slice <N>: <small deliverable>
- Scope:
- Files likely touched:
- Implementation notes:
- Tests:
- Docs/ops updates:
- Rollback plan:
- Definition of Done:
Slices should usually fit in one focused PR/MR.
Phase 5 — Prioritization
Use a lightweight scoring model unless the user provides another one.
Score each candidate 1–5:
| Factor | Meaning |
|---|
| Impact | User/business value if completed. |
| Urgency | Time sensitivity or blocking nature. |
| Confidence | Evidence quality and requirement clarity. |
| Effort inverse | 5 = small/easy, 1 = large/uncertain. |
| Risk reduction | How much this reduces future delivery risk. |
Suggested priority score:
Priority = Impact + Urgency + Confidence + EffortInverse + RiskReduction
Then classify:
Now / P0: must do immediately.
Next / P1: next milestone.
Later / P2: valuable but not blocking.
Park / P3: explicitly defer.
Also identify dependencies: “A must happen before B because …”.
Phase 6 — Roadmap and Execution Plan
Produce a roadmap at three resolutions:
- This week — concrete tasks, validation, expected outputs.
- Next 2–4 weeks — milestone sequence and dependencies.
- Later — strategic backlog and unknowns to resolve.
Each milestone must include:
## Milestone <name>
- Goal:
- User-visible outcome:
- Work slices:
- Validation plan:
- Release/rollback notes:
- Risks and mitigations:
- Exit criteria:
For implementation handoff, create a task plan:
| Task | Priority | Owner/Agent | Dependencies | Validation | Status |
|---|---:|---|---|---|---|
If task-tracking tools are available, create/update tasks for accepted execution work.
Phase 7 — Validation Gates
Every plan must include validation commands or checks.
For this repo, prefer targeted checks:
| Area | Suggested checks |
|---|
| Engine loop/tools/plugins | pnpm --filter pulse-coder-engine test; pnpm --filter pulse-coder-engine typecheck |
| CLI | pnpm --filter pulse-coder-cli test; pnpm --filter pulse-coder-cli build |
| Sandbox | pnpm --filter pulse-sandbox test; pnpm --filter pulse-sandbox typecheck |
| Memory | pnpm --filter pulse-coder-memory-plugin test; pnpm --filter pulse-coder-memory-plugin typecheck |
| Remote server | pnpm --filter @pulse-coder/remote-server build |
| Cross-package change | pnpm run build; targeted tests first, full build last |
| Docs/skill-only change | parse/read file, check frontmatter has name and description; no build required unless loader changed |
If validation cannot run, state why and provide the next best evidence.
Required Output Format
Use this structure by default:
# 项目分析与需求推进方案
## 1. 结论先行
- 当前判断:
- 最应该做的 3 件事:
- 最大风险:
## 2. 证据来源
| Evidence | Path/Command | What it supports | Confidence |
|---|---|---|---|
## 3. 当前项目地图
- 产品定位:
- 架构层次:
- 关键模块:
- 主要流程:
## 4. 健康诊断
| Area | Finding | Severity | Evidence | Recommendation |
|---|---|---:|---|---|
## 5. 需求拆解
<PRD-lite blocks>
## 6. 优先级排序
| Candidate | Impact | Urgency | Confidence | EffortInverse | RiskReduction | Priority | Tier |
|---|---:|---:|---:|---:|---:|---:|---|
## 7. Roadmap
### This week
### Next 2–4 weeks
### Later
## 8. 执行任务板
| Task | Priority | Dependencies | Validation | Status |
|---|---:|---|---|---|
## 9. 开放问题与决策点
- Decision needed:
- Assumption to verify:
## 10. 下一步
- Recommended immediate next action:
- If you want me to execute: <specific first slice>
Quality Checklist
Before finalizing, verify:
Completion Criteria
This skill is complete when the user receives either:
- A repo-wide project analysis plus prioritized roadmap, or
- A focused requirement plan with task-level execution handoff, or
- A clear blocker report explaining what information is missing and the minimum questions needed to proceed.