create-workflow
INVOKE THIS SKILL IMMEDIATELY when user asks to: write/create/build a Nextflow pipeline or workflow, create any bioinformatics pipeline (RNA-seq, DNA-seq, variant calling, ChIP-seq, etc.), or compose/chain Nextflow modules from the Nextflow Registry. This skill handles all Nextflow workflow creation tasks.
nextflow-io · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add nextflow-io/agent-skills --skill create-workflow技能说明
name: create-workflow description: | INVOKE THIS SKILL IMMEDIATELY when user asks to: write/create/build a Nextflow pipeline or workflow, create any bioinformatics pipeline (RNA-seq, DNA-seq, variant calling, ChIP-seq, etc.), or compose/chain Nextflow modules from the Nextflow Registry. This skill handles all Nextflow workflow creation tasks. allowed-tools: Bash, Read, Edit, Write, Glob, Grep, Skill
Nextflow Workflow Writer
Create complete Nextflow workflows by composing validated modules from the Nextflow Registry.
Modules are published under namespaces (e.g. nf-core/fastqc), and these skills compose modules from any of them.
Requires Nextflow 26.04 or later (for the nextflow module commands used during validation).
NEVER write a wrapper workflow just to run/test a single module.
❌ WRONG - Writing a workflow to run one module:
// DO NOT DO THIS - not even when a module run fails!
include { FASTQC } from 'nf-core/fastqc'
workflow { FASTQC(Channel.fromPath('data/*.fq.gz')) }
✅ CORRECT - Use the run-module skill:
Skill(skill="run-module")
The run-module skill:
- Uses
nextflow module search/viewto discover and get proper inputs/parameters - Runs modules directly via
nextflow module run <namespace>/<module> - No wrapper workflow needed
⚠️ When a module run fails due to missing args:
- DO NOT write a wrapper workflow as a "fix"
- Instead: run
nextflow module view <module>to get correct parameters - Fix the command-line arguments and re-run directly
Only write a workflow in Step 4 when composing multiple validated modules together.
4-Step Workflow Creation Process
ALWAYS follow this structured process when creating a new workflow:
Step 1: Identify Modules and Propose Plan
- Use
nextflow module search <term>to find Registry modules for each processing step - Use
nextflow module view <name>to understand inputs/outputs of each module - Present a plan to the user with:
- List of identified modules
- Processing sequence (which module runs first, second, etc.)
- Data flow between modules (outputs → inputs)
STOP and wait for user approval before proceeding.
Step 2: User Agreement
- Wait for user to review and approve the proposed plan
- Address any questions or modifications requested
- Only proceed when user explicitly agrees
Step 3: Validate Modules ONE BY ONE (MANDATORY)
NEVER skip this step. After user agreement:
- Determine appropriate test data for validation
- For EACH module in the plan, sequentially:
- Install and run with test data: Invoke
Skill(skill="run-module") - Verify outputs - confirm expected data is produced
- Only proceed to next module after current one succeeds
- Install and run with test data: Invoke
- Log ALL module run commands and their outputs to a debug file with the
.modules-validation-prefix - If a command fails, stop and show the user the command used and the output generated before trying something else
Note: The
run-moduleskill usesnextflow modulecommands for discovery, configuration, and execution — modules are installed on-the-fly.
DO NOT proceed to Step 4 until ALL modules have been individually validated.
Step 4: Compose Final Workflow
Only after ALL modules run successfully:
-
Configure
nextflow.configwith Wave + Conda:wave.enabled = true wave.strategy = 'conda,container' docker.enabled = true -
Write the workflow script compositing all validated steps using Nextflow managed modules (no
./prefix — Nextflow automatically downloads and installs them from the Nextflow Registry):include { MODULE_A } from 'nf-core/module_a' include { MODULE_B } from 'nf-core/module_b' workflow { MODULE_A(input_ch) MODULE_B(MODULE_A.out.results) } -
Run the complete workflow using the same test data to validate end-to-end
Critical Guidelines
Command Execution
- Use
-resumeflag to leverage cached results when appropriate - Use absolute paths, never relative paths
File Handling
- When specifying multiple files, separate with comma and wrap in double quotes:
--input "file1.fq,file2.fq" - ALWAYS expand wildcards/globs to comma-separated file lists before running
- Reference task IDs from stdout to locate output files in work directories
Module Include Syntax
include { MOD } from 'nf-core/module'— Nextflow managed module (default). Nextflow automatically downloads and installs it from the Nextflow Registry. Always prefer this form. (nf-corehere is the namespace; substitute the actual namespace of the module you are using.)include { MOD } from './modules/nf-core/module/main.nf'— Local file path, resolved against the working directory. Only use when referencing locally modified modules.
Module Selection
- Prefer single-tool modules over sub-workflows
- Do not write wrapper workflows to test single modules - use
Skill(skill="run-module")instead - Use
nextflow module searchto find modules, thennextflow module viewfor details
Debugging Protocol
- Check the task work directory using the task ID from stdout
- Examine
.command.log,.command.err, and.command.outfiles - Verify input files exist and are accessible
- Check resource requirements (memory, CPUs) match available resources
Skill Delegation (IMPORTANT)
Delegate module-specific tasks to specialized skills using the Skill tool:
| Task | Invoke Skill |
|---|---|
| Install/run/test a module | Skill(skill="run-module") |
When to Delegate
- Step 3 (Validate Modules): Use
run-moduleskill for each module validation
Example: Step 3 Validation
For each module in your plan:
1. Invoke: Skill(skill="run-module") → install, run with test data, and verify outputs
2. Only proceed to next module after success
These skills contain detaile
阅读完整来源 ↗