version: 1.0.0 author: Claude Code tags: [lark, testing, automation, markdown, token-efficient, mcp]

Lark Agent Simple Skill

Token-efficient Lark task creation from markdown test files with 60-70% token reduction compared to the original lark-agent.

Overview

The Lark Agent Simple skill provides an optimized workflow for:

•Parsing markdown test files into minimal JSON (data only)
•Executing Lark MCP calls directly (no workflow generation)
•Creating hierarchical Lark tasks (3-level structure)
•Reporting results with summary

This skill eliminates the token overhead of workflow generation by moving execution logic from Python to the slash command.

Architecture

Token Efficiency Design

Traditional Approach (lark-agent):

code

Markdown → Python → Workflow Plan JSON (2000-5000 tokens)
→ Claude interprets plan → Claude executes MCP
Total: 10,000-25,000 tokens

Optimized Approach (lark-agent-simple):

code

Markdown → Python → Minimal Data JSON (500-1000 tokens)
→ Claude executes MCP directly
Total: 3,000-8,000 tokens

Savings: 60-70% token reduction

Component Roles

•
Python Parser (lark_agent_simple.py)
- •Role: Data transformation only (markdown → JSON)
- •Output: Minimal structured data (no workflow instructions)
- •Size: ~150 lines (vs 250+ in original)
•
Slash Command (.claude/commands/lark-agent-simple.md)
- •Role: Execution logic
- •Contains: Direct MCP call instructions
- •Advantage: Claude executes immediately (no interpretation)
•
SKILL.md (this file)
- •Role: Documentation and activation guide
- •Contains: Usage instructions, examples, reference

When to Use This Skill

Activate this skill when user:

•Wants to create Lark tasks from markdown test files
•Needs token-efficient execution
•Mentions "lark agent simple" or "efficient lark"
•Has existing test documentation in markdown format
•Wants to avoid token overhead of workflow generation

Usage

Command

bash

/lark-agent-simple <markdown-file> [--owner="Name"] [--due-date="YYYY-MM-DD"]

Parameters

•<markdown-file> (required): Path to markdown test file
•--owner (optional): Task owner name (default: "Test User")
•--due-date (optional): Target date YYYY-MM-DD (default: 14 days from now)
•--start-date (optional): Start date YYYY-MM-DD (default: today)

Examples

bash

# Basic usage
/lark-agent-simple tests/manual/onboarding-test.md

# With custom owner and due date
/lark-agent-simple tests/manual/login-test.md --owner="QA Team" --due-date="2025-12-31"

# With start and due dates
/lark-agent-simple tests/manual/api-test.md --owner="Dev Team" --start-date="2025-10-20" --due-date="2025-11-03"

Markdown Format

Your test file must follow this structure:

markdown

# Test Title
Test description

## Test Scenario: Scenario Name
Scenario description

### Task: Task Name
1. Step one
2. Step two
3. Step three

Expected Result: What should happen

Format Requirements

•H1 (#): Test title (required)
•H2 (##): Test scenarios (starts with "Test Scenario:")
•H3 (###): Individual tasks (starts with "Task:")
•Steps: Numbered list items
•Expected Result: Line starting with "Expected Result:"

Output

Compact JSON Data

The Python parser outputs minimal JSON:

json

{
  "success": true,
  "data": {
    "test": {
      "title": "Test Title",
      "description": "Test description",
      "owner": "QA Team",
      "start_date": "2025-10-19",
      "due_date": "2025-11-02"
    },
    "scenarios": [
      {
        "id": "scenario-0-1729300000000",
        "title": "Scenario Title",
        "description": "Scenario description",
        "tasks": [
          {
            "id": "task-0-0-1729300000000",
            "title": "Task Title",
            "description": "Steps...",
            "expected_result": "Expected outcome"
          }
        ]
      }
    ],
    "metadata": {
      "total_scenarios": 3,
      "total_tasks": 10,
      "source_file": "path/to/test.md"
    }
  }
}

Lark Task Structure

Creates 3-level hierarchy:

•Level 1: Parent task (test overview)
•Level 2: Scenario tasks (marked as milestones if they have subtasks)
•Level 3: Individual test tasks

Integration with Lark MCP

This skill uses the following Lark MCP tools:

•task_v2_tasklist_create - Create task list
•task_v2_task_create - Create parent task
•task_v2_taskSubtask_create - Create scenario and individual tasks

All calls use useUAT: true for user access token authentication.

Workflow Details

The slash command executes this workflow:

•Parse Markdown → Execute Python script → Get compact JSON
•Create Tasklist → Call MCP → Save tasklistGuid
•Create Parent Task → Call MCP → Save parentTaskGuid
•Create Scenarios → Loop + Call MCP → Save scenarioGuids[id]
•Create Tasks → Loop + Call MCP → Save taskGuids[id]
•Report Summary → Display counts and IDs

Error Handling

The parser validates:

•File exists and is markdown (.md or .markdown)
•File has valid structure (H1, H2, H3 hierarchy)
•Dates are in correct format

Errors are returned as:

json

{
  "success": false,
  "error": "Error message",
  "error_type": "FileNotFoundError"
}

Key Benefits

Token Efficiency

•60-70% reduction in token usage
•Minimal data transfer (no workflow instructions)
•Direct execution (no interpretation overhead)

Speed

•Faster execution (fewer round trips)
•No workflow generation delay
•Immediate MCP calls

Maintainability

•Clean separation (data vs logic)
•Logic in one place (slash command)
•Easy to update execution pattern

Simplicity

•Straightforward data flow
•No complex abstraction layers
•Clear execution path

Comparison with Original lark-agent

Aspect	lark-agent (Original)	lark-agent-simple (New)
Python Output	Full workflow plan (2000-5000 tokens)	Minimal data (500-1000 tokens)
Execution Logic	In Python (generates instructions)	In slash command (direct)
MCP Calls	Claude interprets from plan	Claude executes directly
Token Usage	10,000-25,000 per execution	3,000-8,000 per execution
Python Lines	~250+ lines	~150 lines
Complexity	High (abstraction layers)	Low (direct flow)
Speed	Slower (interpretation)	Faster (direct)

Files

code

.claude/skills/lark-agent-simple/
├── SKILL.md                          # This file (documentation)
├── run.py                            # Entry point
├── scripts/
│   ├── markdown_parser.py            # Markdown parser (copied from original)
│   └── lark_agent_simple.py          # Minimal parser (data only)
└── examples/
    └── sample-test.md                # Example test file

.claude/commands/
└── lark-agent-simple.md              # Slash command (execution logic)

Testing

To test the parser independently:

bash

cd .claude/skills/lark-agent-simple
python run.py examples/sample-test.md --owner="Test User" --due-date="2025-12-31"

This outputs compact JSON that can be verified before running the full workflow.

Troubleshooting

Parser Issues

If parsing fails:

•Check markdown file structure (H1 > H2 > H3 hierarchy)
•Verify file encoding is UTF-8
•Ensure scenario headers start with "Test Scenario:"
•Ensure task headers start with "Task:"

MCP Issues

If Lark task creation fails:

•Verify Lark MCP server is running
•Check user has permissions in Lark
•Validate date formats (YYYY-MM-DD)
•Review error messages from MCP

Date Issues

If dates are invalid:

•Use YYYY-MM-DD format only
•Ensure due date is after start date
•Check dates are in the future (if required by Lark)

Version History

v1.0.0 (Initial Release)

•Minimal JSON parser (data only)
•Direct MCP execution via slash command
•60-70% token reduction
•3-level task hierarchy
•Compact output format

Support

For issues or questions:

•Review this documentation
•Check example files in examples/
•Verify markdown format matches requirements
•Test parser independently before full execution

Remember: This skill is designed for token efficiency. If you need detailed workflow tracking or complex verification, consider using the original lark-agent skill instead.