@markus-global/cli 0.8.4 → 0.8.5-rc.0

package/templates/skills/feishu-interaction/SKILL.md

@@ -0,0 +1,103 @@
+---
+name: feishu-interaction
+description: Feishu/Lark platform interaction via MCP - documents, tasks, calendar, Bitable, messaging
+---
+
+# Feishu/Lark Platform Interaction
+
+You have access to the Feishu/Lark platform via MCP tools (prefixed `feishu-lark__`).
+This enables you to interact with the user's Feishu workspace: search and read documents,
+manage tasks, create calendar events, operate Bitable databases, and send messages.
+
+## Available Capabilities
+
+### Documents (文档)
+- `docx.v1.document.rawContent` — Read document content
+- `docx.builtin.import` — Import documents (create new docs from content)
+- `docx.builtin.search` — Search documents by keyword
+- `wiki.v2.space.getNode` — Get Wiki node content
+- `wiki.v1.node.search` — Search Wiki nodes
+
+### Tasks (任务)
+- `task.v2.task.create` — Create a new task with title, description, due date
+- `task.v2.task.patch` — Update an existing task
+- `task.v2.task.addMembers` — Add members to a task
+- `task.v2.task.addReminders` — Add reminders to a task
+
+### Calendar (日历)
+- `calendar.v4.calendarEvent.create` — Create a calendar event
+- `calendar.v4.calendarEvent.patch` — Modify a calendar event
+- `calendar.v4.calendarEvent.get` — Get calendar event details
+- `calendar.v4.freebusy.list` — Query free/busy status
+- `calendar.v4.calendar.primary` — Get primary calendar info
+
+### Bitable (多维表格)
+- `bitable.v1.appTable.list` — List tables in a base
+- `bitable.v1.appTableField.list` — List fields in a table
+- `bitable.v1.appTableRecord.search` — Search records
+- `bitable.v1.appTableRecord.create` — Create records
+- `bitable.v1.appTableRecord.update` — Update records
+
+### Messaging (消息)
+- `im.v1.message.create` — Send a message to a chat or user
+- `im.v1.message.list` — List messages in a chat
+- `im.v1.chat.create` — Create a new group chat
+- `im.v1.chat.list` — List chats the bot is in
+
+### Contacts (通讯录)
+- `contact.v3.user.batchGetId` — Batch get user IDs by email/mobile
+
+## Best Practices
+
+### Document Operations
+1. **Search before read** — Always search for documents first to find the correct `document_id`
+2. **Prefer Wiki nodes** — If the user mentions "知识库" or "Wiki", use wiki APIs
+3. **Cite sources** — When referencing document content in replies, include the document title
+
+### Task Management
+1. **Match user intent** — When user says "帮我创建一个任务", use `task.v2.task.create`
+2. **Set due dates** — Always ask for or infer a due date when creating tasks
+3. **Add members** — If the user mentions specific people, resolve their user IDs via contacts API first
+
+### Calendar Events
+1. **Check availability** — Before creating an event, use `freebusy.list` to check conflicts
+2. **Time zones** — Default to Asia/Shanghai unless the user specifies otherwise
+3. **Include details** — Set description, location, and attendees when provided
+
+### Bitable Operations
+1. **Discover structure** — List tables and fields before searching records
+2. **Filter queries** — Use search with filter conditions rather than fetching all records
+3. **Batch operations** — For multiple records, prefer batch create/update when available
+
+### Messaging
+1. **Format appropriately** — Use markdown formatting for rich messages
+2. **Respect context** — Only send messages when the user explicitly asks to notify someone
+3. **Interactive cards** — For structured information, use `msg_type: 'interactive'` with card JSON
+
+## Important Notes
+
+- Tool names are prefixed with `feishu-lark__` (e.g., `feishu-lark__docx.builtin.search`)
+- Document editing is NOT supported — you can only read and import
+- File upload/download is NOT supported via MCP
+- Always handle API errors gracefully and report them to the user
+- Respect rate limits: 1000 requests/minute per API endpoint
+
+## Permission Error Handling
+
+When a tool call returns an error containing `code: 99991672` or "Access denied...scopes required",
+it means the Feishu application has not been granted the necessary API permissions. In this case:
+
+1. **Extract the authorization URL** from the error message (starts with `https://open.feishu.cn/app/`)
+2. **Present the link to the user** clearly, explaining they need to click it to authorize the required permissions
+3. **Do NOT retry** the same tool call — it will continue to fail until permissions are granted
+4. **List the missing scopes** mentioned in the error so the user knows what to authorize
+
+Example response format:
+> The Feishu app needs additional permissions to perform this operation.
+> Please open the following link to authorize, then try again:
+>
+> [https://open.feishu.cn/app/cli_xxx/auth?q=docs:doc,drive:drive,...](https://open.feishu.cn/app/cli_xxx/auth?q=docs:doc,drive:drive,...)
+>
+> Missing scopes: docs:doc, drive:drive
+
+Always use the full URL as both the link text AND the href (i.e. `[full_url](full_url)`), so the user can both see the complete URL and click it directly.

package/templates/skills/feishu-interaction/skill.json

@@ -0,0 +1,26 @@
+{
+  "type": "skill",
+  "name": "feishu-interaction",
+  "displayName": "Feishu Interaction",
+  "version": "1.0.1",
+  "description": "Interact with Feishu/Lark platform - search documents, manage tasks, create calendar events, operate Bitable, and send messages",
+  "author": "markus",
+  "category": "integration",
+  "tags": ["feishu", "lark", "integration", "document", "calendar", "task", "mcp"],
+  "i18n": {
+    "zh-CN": {
+      "displayName": "飞书交互",
+      "description": "与飞书平台交互 - 搜索文档、管理任务、创建日历事件、操作多维表格、发送消息"
+    }
+  },
+  "skill": {
+    "skillFile": "SKILL.md",
+    "requiredPermissions": ["network"],
+    "isolation": "shared",
+    "mcpServers": {
+      "feishu-lark": {
+        "reuseGlobal": true
+      }
+    }
+  }
+}

package/templates/skills/self-evolution/SKILL.md

@@ -224,6 +224,37 @@ During heartbeat, check your revision rate:
 - Consider: would a ROLE.md rule or a HEARTBEAT.md check have prevented any recent failures?
 - Escalate recurring mistakes: insight → MEMORY.md procedure → ROLE.md rule or HEARTBEAT.md check
 
+## Team-Wide Pattern Recognition
+
+When you notice patterns that affect the whole team — not just yourself:
+
+| Signal | Action |
+|--------|--------|
+| Same bug type recurring across multiple agents | Create a skill with prevention guidelines |
+| Common tool usage mistakes | Add to team NORMS.md or propose a skill |
+| Repeated review feedback across the team | Propose a coding standard update |
+| Process bottleneck affecting multiple tasks | Flag to manager with data |
+
+Use `deliverable_create` to share patterns as team knowledge. Individual insights go to `memory_save`; team patterns deserve broader visibility.
+
+## The Ratchet Principle
+
+Apply the keep-or-discard discipline to your own evolution:
+- When you try a new approach and it works better, commit it to knowledge (`memory_save`, MEMORY.md, or ROLE.md).
+- When an approach fails, record WHY it failed and move on — don't keep trying the same thing.
+- Your ROLE.md, HEARTBEAT.md, and MEMORY.md are your "program.md" — they are the instructions that determine your effectiveness. Improving them IS your self-evolution.
+
+### Delete the Harness
+
+Your accumulated knowledge exists to compensate for your limitations. As you improve — through practice, better tools, or model upgrades — some of that knowledge becomes overhead:
+
+- Periodically re-read your MEMORY.md and HEARTBEAT.md. If a procedure or check is now something you do naturally (or the platform handles automatically), remove it.
+- A growing MEMORY.md is not always a sign of progress. Prune entries that are no longer needed — the harness that grows monotonically is a harness you have stopped reading.
+- After platform updates or model improvements, audit your procedures: delete anything the system now does for free.
+- The goal is a lean, high-signal knowledge base — not an exhaustive archive.
+
+---
+
 ## Rules
 
 - **DO** save insights immediately while context is fresh

package/templates/teams/content-team/NORMS.md

@@ -48,3 +48,20 @@
 - 所有内容成品通过 `deliverable_create` 登记，确保可追溯。
 - 紧急发布需求使用 `notify_user` 通知编辑。
 - 各平台数据表现定期通过 `task_note` 汇总。
+
+## 质量门禁
+
+每个内容环节的交接需通过对应的质量检查：
+
+| 环节 | 质量要求 |
+|------|---------|
+| 选题 → 创作 | 选题有明确角度、目标平台、受众定义 |
+| 创作 → 审核 | 初稿完整、事实可追溯、符合平台格式 |
+| 审核 → 发布 | 编辑确认准确性、调性一致性、合规性 |
+| 发布 → 复盘 | 发布链接已记录、数据追踪已配置 |
+
+## 错误恢复
+
+- **事实错误**：发现后立即通知编辑，评估是否需要更正或撤回。在 `task_note` 中记录错误原因和修正方案。
+- **平台违规**：内容被平台警告或删除时，分析原因，更新内容规范，避免再犯。
+- **数据异常**：发布后数据表现远低于预期时，分析选题、标题、发布时间等因素，记录到团队知识库。

package/templates/teams/dev-squad/NORMS.md

@@ -60,3 +60,29 @@ This is the most important rule for parallel development:
 - Document non-obvious decisions as `deliverable_create` (type: "architecture_decision").
 - Save reusable patterns and gotchas via `memory_save` with appropriate tags.
 - After completing a complex task, record lessons learned during the self-evolution reflection.
+
+## Quality Gates by Phase
+
+Every phase transition requires passing its quality gate:
+
+| Transition | Gate |
+|------------|------|
+| Plan → Implement | Tasks have clear acceptance criteria, file ownership defined, dependencies mapped |
+| Implement → Review | Tests pass, changes within scope, no debug artifacts, summary in task notes |
+| Review → Complete | Reviewer verified correctness, conventions, security, and scope. Branch merged. |
+| Any → Blocked | Blocker documented with what was tried and why it failed |
+
+## Error Recovery
+
+- **Build failures**: Developer fixes immediately. If systemic (dependency issue), Tech Lead coordinates.
+- **Merge conflicts**: Developer resolves in their worktree. If conflict involves another developer's files, coordinate via `agent_send_message` before resolving.
+- **Review rejection**: Developer addresses ALL feedback items — don't skip any. Re-submit when done.
+- **Stale branches**: If a branch falls behind main by >10 commits, rebase before submitting for review.
+
+## Security Checklist
+
+Before approving any task that touches these areas, the reviewer must verify:
+- **Authentication/Authorization**: Proper token validation, permission checks, session management
+- **User Input**: Input validation, parameterized queries, output encoding
+- **Secrets**: No hardcoded credentials, proper secret management
+- **Dependencies**: No known CVEs in new dependencies

package/templates/teams/dev-squad/team.json

@@ -35,28 +35,28 @@
         "role": "worker",
         "roleName": "developer",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       },
       {
         "name": "Frontend Developer",
         "role": "worker",
         "roleName": "developer",
         "count": 1,
-        "skills": ["chrome-devtools", "self-evolution"]
+        "skills": ["chrome-devtools", "self-evolution", "coding-tools"]
       },
       {
         "name": "Code Reviewer",
         "role": "worker",
         "roleName": "reviewer",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       },
       {
         "name": "QA Engineer",
         "role": "worker",
         "roleName": "qa-engineer",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       }
     ]
   }

package/templates/teams/engineering-pod/NORMS.md

@@ -76,3 +76,36 @@ Edit anything in the "Shared" column only after notifying the team via `agent_se
 - Frontend components must handle loading, error, and empty states.
 - Infra changes must be idempotent and rollback-safe.
 - Security-critical code (auth, permissions, data access) requires explicit review notes.
+
+## State Machine: Phase Transitions
+
+```
+DESIGN ──[design brief approved]──> CONTRACT
+CONTRACT ──[all contracts published]──> IMPLEMENT
+IMPLEMENT ──[all subtasks done, tests pass]──> INTEGRATE
+INTEGRATE ──[contracts verified]──> REVIEW
+REVIEW ──[approved + merged]──> DEPLOY
+DEPLOY ──[smoke tests pass]──> DONE
+
+On failure at any phase:
+  IMPLEMENT ──[build fails]──> IMPLEMENT (fix and retry)
+  REVIEW ──[rejected]──> IMPLEMENT (address feedback)
+  DEPLOY ──[smoke fails]──> ROLLBACK ──> IMPLEMENT
+```
+
+## Integration Quality Gates
+
+Before moving from IMPLEMENT to INTEGRATE:
+- Each layer passes its own test suite independently
+- API responses match published contract schemas
+- No breaking changes to shared interfaces without team-wide notification
+- Cross-layer error handling verified (what happens when Backend returns 500 to Frontend?)
+
+## Error Recovery Coordination
+
+| Failure | Owner | Action |
+|---------|-------|--------|
+| Build failure in one layer | That layer's engineer | Fix and re-run; notify team if shared dependency |
+| Contract violation | Architect | Mediate between layers; update contract if needed |
+| Integration test failure | Architect + affected engineers | Identify which layer deviated; coordinate fix |
+| Deployment failure | Infra Engineer | Rollback; create post-mortem task |

package/templates/teams/engineering-pod/team.json

@@ -35,28 +35,28 @@
         "role": "worker",
         "roleName": "developer",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       },
       {
         "name": "Frontend Engineer",
         "role": "worker",
         "roleName": "developer",
         "count": 1,
-        "skills": ["chrome-devtools", "self-evolution"]
+        "skills": ["chrome-devtools", "self-evolution", "coding-tools"]
       },
       {
         "name": "Infra Engineer",
         "role": "worker",
         "roleName": "devops",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       },
       {
         "name": "Senior Reviewer",
         "role": "worker",
         "roleName": "reviewer",
         "count": 1,
-        "skills": ["self-evolution"]
+        "skills": ["self-evolution", "coding-tools"]
       }
     ]
   }

package/templates/teams/research-lab/NORMS.md

@@ -75,3 +75,18 @@
 - **建设性异议**："我发现了与 X 矛盾的证据，因为..."而非"X 错了。"
 - **跟踪置信度**：使用显式级别（高/中/低），随证据变化更新。
 - 协调秘书确保关键沟通内容记录可追溯。
+
+## 质量门禁
+
+| 阶段 | 质量要求 |
+|------|---------|
+| 框架 → 调研 | 研究问题精确、成功标准明确、假设可检验 |
+| 调研 → 验证 | 每个发现有证据支持、置信度已标注、负证据已记录 |
+| 验证 → 综合 | 所有交叉审查完成、分歧已解决或升级、质疑已回应 |
+| 综合 → 交付 | 执行摘要完整、方法论透明、建议可操作 |
+
+## 错误恢复
+
+- **研究方向错误**：及时止损，记录为"死胡同"发现（`memory_save` with tags: `["insight", "dead-end"]`），转向下一个假设。
+- **来源冲突**：两个权威来源给出矛盾信息时，深入查证第三方来源，在报告中如实反映不确定性。
+- **时间压力**：如果调研时间不足以完成所有角度，优先完成最高影响的调研线索，明确标注哪些角度未覆盖。

package/templates/teams/startup-team/NORMS.md

@@ -57,3 +57,20 @@ Adjust quality based on what you're building:
 - Save every experiment result via `memory_save` with `tags: ["experiment", "hypothesis"]`.
 - Document architecture decisions that constrain future work via `deliverable_create`.
 - Share customer insights with the whole team — everyone should know what users want.
+
+## Quality Calibration by Risk
+
+| Category | Quality Level | Testing | Review |
+|----------|--------------|---------|--------|
+| Core product (revenue path) | High | Unit + integration tests required | Full review |
+| Growth experiments | Medium | Happy path test, manual verification | Quick review |
+| Internal tools | Medium | Basic tests, error handling | Self-review acceptable |
+| One-off analysis | Low | Verify results make sense | No formal review |
+
+Adjust quality investment based on what you're building. Over-engineering an experiment wastes time. Under-testing core product creates debt.
+
+## Error Recovery
+
+- **Experiment fails**: Document what was tested and what was learned. Move on — don't fix failed experiments, learn from them.
+- **Core product bug**: Fix immediately. Write regression test. Do post-mortem if user-facing.
+- **Deployment issue**: Rollback first, investigate second. Never debug in production.

package/templates/teams/startup-team/team.json

@@ -33,7 +33,7 @@
         "role": "worker",
         "roleName": "developer",
         "count": 2,
-        "skills": ["chrome-devtools", "self-evolution"]
+        "skills": ["chrome-devtools", "self-evolution", "coding-tools"]
       },
       {
         "name": "Growth Lead",