Task Decomposer & Skill Generator

This skill helps decompose complex user requests into executable subtasks, identify required capabilities for each task, search for existing skills from the open skills ecosystem, and automatically create new skills when no existing solution is available.

Core Workflow

CODEBLOCK0

Phase 1: Task Analysis & Decomposition

When receiving a user request, follow these steps:

Step 1: Understand User Intent

Analyze the request to identify:

• - Core objective: What is the end goal?
• Domains involved: What areas of expertise are needed?
• Trigger mechanism: One-time, scheduled, or event-driven?

Example analysis:
CODEBLOCK1

Step 2: Decompose into Atomic Tasks

Break down the complex task into minimal executable units:

CODEBLOCK2

Phase 2: Capability Identification

Map each subtask to a capability type from the universal capability taxonomy.

Universal Capability Types

Capability	Description	Search Keywords
INLINECODE0	Web navigation, interaction, scraping	browser, selenium, puppeteer, playwright, scrape
INLINECODE1

Internet search and information retrieval | search, google, bing, duckduckgo | | api_integration | Third-party API communication | api, rest, graphql, webhook, {service-name} | | data_extraction | Parse and extract structured data | parse, extract, scrape, ocr, pdf | | data_transformation | Convert, clean, transform data | transform, convert, format, clean, etl | | content_generation | Create text, images, or other content | generate, write, create, summarize, translate | | file_operations | Read, write, manipulate files | file, read, write, csv, excel, json, pdf | | message_delivery | Send notifications or messages | notify, send, email, slack, discord, telegram | | scheduling | Time-based task execution | schedule, cron, timer, daily, weekly | | authentication | Identity and access management | auth, oauth, login, token, credentials | | database_operations | Database CRUD operations | database, sql, mongodb, query, store | | code_execution | Run scripts or programs | execute, run, script, shell, python | | version_control | Git and code repository operations | git, github, gitlab, commit, pr, review | | testing | Automated testing and QA | test, jest, pytest, e2e, unit | | deployment | Application deployment and CI/CD | deploy, docker, kubernetes, ci-cd, release | | monitoring | System and application monitoring | monitor, alert, log, metrics, health |

Capability Identification Process

For each subtask:

1. 1. Analyze the task description and requirements
2. Match to one or more capability types
3. Generate search keywords for skill discovery

Example:
CODEBLOCK3

Phase 3: Skill Search

Use the Skills CLI to search for existing skills at https://skills.sh/

Search Process

For each capability need, search using relevant keywords:

CODEBLOCK4

Evaluate Search Results

When results are returned:
CODEBLOCK5

Evaluate each result for:

• - Relevance: Does it match the required capability?
• Completeness: Does it cover all needed functionality?
• Quality: Is it well-documented and maintained?

Generate Capability Mapping

CODEBLOCK6

Phase 4: Gap Analysis

Identify tasks without matching skills:

Built-in Capabilities (No Skill Needed)

These capabilities are typically handled by the agent's native abilities:

• - content_generation - LLM's native text generation
• INLINECODE17 - Basic data manipulation via code
• INLINECODE18 - Direct script execution
• INLINECODE19 - System-level cron/scheduler configuration

Skills Required

For capabilities without built-in support, determine:

1. 1. Skill exists: Install from skills.sh
2. Skill not found: Create new skill

Phase 5: Skill Creation

When no existing skill matches a required capability, create a new skill.

Skill Creation Process

1. 1. Define scope: Determine what the skill should do
2. Design interface: Define inputs, outputs, and usage patterns
3. Create SKILL.md: Write the skill definition file
4. Add resources: Include scripts, references, or assets as needed

Skill Template

CODEBLOCK7bash
{Basic command or code example}
CODEBLOCK8

Initialize New Skill

CODEBLOCK9

Phase 6: Generate Execution Plan

Compile all information into a structured execution plan:

CODEBLOCK10

Task Decomposition Principles

Principle 1: Atomicity

Each subtask should be the minimal executable unit with clear input and output.

Principle 2: Independence

Minimize dependencies between tasks to allow parallel execution where possible.

Principle 3: Verifiability

Each task should have a clear way to verify successful completion.

Principle 4: Reusability

Identify reusable patterns and prefer creating general-purpose skills.

Principle 5: Single Responsibility

Each task should do one thing well.

Output Format

Present the decomposition results in a structured format:

CODEBLOCK11

Examples

Example 1: Workflow Automation

User Request:
CODEBLOCK12

Decomposition:
CODEBLOCK13

Example 2: Data Pipeline

User Request:
CODEBLOCK14

Decomposition:
CODEBLOCK15

Best Practices

1. 1. Start with skill search: Always check https://skills.sh/ before creating new skills
2. Use specific search terms: Combine capability keywords with domain terms
3. Leverage built-in capabilities: Don't create skills for things the agent can do natively
4. Create reusable skills: Design new skills to be general-purpose when possible
5. Document thoroughly: New skills should have clear usage instructions
6. Verify before proceeding: Confirm skill installation before executing tasks
7. Handle errors gracefully: Include fallback strategies in execution plans

Integration with find-skills

This skill works in conjunction with the find-skills skill for discovering existing solutions:

CODEBLOCK16

Notes

• - Always search for existing skills before creating new ones
• Built-in capabilities (LLM, basic code) don't require skills
• Skill creation requires user confirmation before proceeding
• Complex workflows may need multiple skills working together