AgentSkillsCN

vibe-kanban-task-execution

执行特定的 Vibe Kanban 任务。从会话启动、编码、测试，到任务完成，全程专注于单个任务的高效推进。

SKILL.md
--- frontmatter
name: vibe-kanban-task-execution
description: Execute a specific Vibe Kanban task. Handles session start, coding, testing, and completion for single-task focused work.

Vibe Execution Skill

This skill provides a streamlined workflow for executing a single Vibe Kanban task from start to finish. It focuses on the "Worker" role: head-down implementation of a fine-grained, atomic task.

When to Use

  • •When you are assigned a specific task_id or asked to "start working on task X".
  • •When you need to execute a single, well-defined task from the board.
  • •During the "Implementation Loop" phase of the development workflow.

Instructions

1. Task Discovery & Session Awareness

Before starting work, retrieve the task details and workspace context:

  1. •Session Awareness: If you are already running within a Vibe Kanban workspace session (e.g., as a coding agent started via start_workspace_session), call get_context first.
    • •Use the returned metadata to instantly identify your project_id, task_id, and repo_ids.
    • •This reduces token waste and redundant searches.
  2. •Task Details:
    • •Use get_task with your task_id to retrieve full details.
    • •Read the task description to understand:
      • •Acceptance criteria
      • •Test plan
      • •Dependencies (if any)
      • •Parent task reference (if subtask)
  3. •Fallback Discovery: If get_context is not available or more info is needed, use list_projects and list_tasks to find available work.

2. Context Aggregation

CRITICAL: Before starting execution, aggregate full context from parent tasks and related work.

  1. •

    Read Current Task Description: Parse structured sections (Context, Plan, Related Tasks).

  2. •

    Read Parent Context (if subtask):

    • •Check task description for Parent Task: #[id] reference
    • •Use get_task to fetch parent task details
    • •Extract ## Context section from parent task
    • •Note: Parent context contains architectural decisions, constraints, goals, and rationale
  3. •

    Check Related Tasks for surrounding context:

    • •Use list_tasks with same project_id to see all tasks in the project
    • •Filter by status to understand project state:
      • •done: Completed work that may affect your task
      • •inprogress: Work happening in parallel
      • •todo: Upcoming work that may depend on your task
    • •Read descriptions of related tasks (from "Related Tasks" section) if needed
    • •Understand dependencies and relationships
  4. •

    Aggregate Context:

    • •Full Context = Parent Context (if exists) + Current Task Plan + Related Task Status
    • •Use this aggregated context to understand:
      • •Architectural decisions that guide your work
      • •Constraints you must work within
      • •Related systems you'll interact with
      • •Dependencies and relationships
      • •Overall project state
  5. •

    Document Context Understanding: Before starting, briefly note key context points that will guide your work.

3. Infrastructure Discovery

Identify the repository context and automation scripts:

  1. •Repo Metadata: If get_context was used, you already have the repo_ids. Otherwise, call list_repos with the project_id to identify the repositories involved.
  2. •Environment Scripts: Use get_repo to inspect setup, cleanup, and dev server scripts if environment configuration is required.

4. Start Execution

Begin working on the task:

  1. •Update Status: Use update_task to set status to inprogress.
  2. •Start Session: Call start_workspace_session with:
    • •task_id: The task you're working on.
    • •executor: The coding agent executor (e.g., CURSOR_AGENT, CLAUDE_CODE, AMP, GEMINI, CODEX, OPENCODE, QWEN_CODE, COPILOT, DROID). See Supported Executors for the full list.
    • •repos: An array of objects with repo_id and base_branch (usually main or master).

5. Implementation & Logging

During implementation, maintain traceability:

  • •Progress Updates: Use update_task to append execution logs to the task description. Format as ## Execution Log with timestamps and milestones.
  • •Status Transitions: Update status as work progresses:
    • •inprogress: Actively working
    • •inreview: Ready for review (if applicable)

Example: Appending Execution Logs

markdown
## Execution Log

### 2026-01-29 14:30

- Started implementation of validation logic.
- Ran `make test`: 2 tests failing (expected).

### 2026-01-29 15:00

- Fixed edge case in type validation.
- Ran `make test`: All tests passing.
- Linter checks: No violations.

6. Verification & Testing

Before marking complete, ensure quality:

  1. •Run Tests: Execute relevant test commands (e.g., make test).
  2. •Lint Check: Use read_lints or project-specific lint commands.
  3. •Append Results: Update the task description with test results and verification status.

7. Completion

Close the task loop:

  1. •Record Change: Use the manage-changelog skill to record the change fragment (feat, fix, etc.).
  2. •Final Update: Use update_task to:
    • •Set status to done (or inreview if review is required).
    • •Append final execution summary.
    • •Include a link to the changelog fragment if created.

IMPORTANT: You MUST record the change fragment before moving the task to done or inreview.

Single-Task Focus

This skill is designed for atomic, single-task execution. If a task reveals additional work:

  • •Small Additions: Update the current task description to include the new scope.
  • •Large Additions: Create a new task using vibe-kanban-management skill and link it in the current task description.
  • •Dependencies: Document dependencies in task descriptions and coordinate with vibe-kanban-management skill.

Workflow Integration

This skill is part of the execution phase:

  1. •Input: A task_id from the Vibe Kanban board (created by vibe-kanban-management skill).
  2. •Process: Execute the task following the workflow above.
  3. •Output: Completed task with change fragment recorded.

Context Reading Pattern

When executing a subtask, follow this pattern to aggregate full context:

  1. •Read Subtask Description: Parse structured sections
  2. •Extract Parent Reference: Look for Parent Task: #[id] in description
  3. •Fetch Parent Task: Use get_task with parent task ID
  4. •Extract Parent Context: Read ## Context section from parent
  5. •Merge Context: Combine Parent Context + Subtask Plan = Full Understanding
  6. •Check Related Tasks: Use list_tasks to see surrounding state
  7. •Begin Execution: Start work with full context awareness

This pattern ensures you understand the "why" (from parent context) and the "what" (from subtask plan) before implementing.

References

  • •See references/hierarchy.md for an explanation of how Projects, Repos, and Tasks relate.
  • •See docs/spec/agent_design.md for detailed context aggregation patterns.
  • •See references/mcp-tools.md for complete MCP tool documentation.
  • •See .cursor/rules/development_workflow.mdc for the full lifecycle policy.
  • •See .cursor/agents/vibe-worker.md for subagent usage patterns.