"Guides MaaNTE Pipeline JSON, task option JSON, i18n, SceneManager,
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill pipeline-development --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Pipeline Development?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-pipeline-development)More formats (shields.io, HTML) on the badges page.
---
name: pipeline-development
description: "Guides MaaNTE Pipeline JSON, task option JSON, i18n, SceneManager,
controller restrictions, and registry changes."
disable-model-invocation: true
metadata:
disco-role: operating
license: AGPL 3.0
---
# Pipeline Development
## Use This When
Use this sub-skill when the task involves `assets/resource/base/pipeline/**/*.json`, `assets/resource/tasks/*.json`, `assets/interface.json`, UI i18n files, SceneManager public interfaces, or Pipeline recognition/action flow design.
Typical prompts:
- Add a new MaaNTE task or task option.
- Modify a Pipeline node, ROI, recognition threshold, action, `next`, `on_error`, or `pipeline_override`.
- Debug why a task option does not affect a node.
- Register a task in the interface import list.
- Update labels/descriptions across five locale files.
- Add a scene-navigation interface or avoid direct private SceneManager usage.
## Read First
1. [references/pipeline-patterns.md](references/pipeline-patterns.md) for Pipeline JSON style, recognition/action patterns, flow control, delays, and SceneManager rules.
2. [references/task-config-and-i18n.md](references/task-config-and-i18n.md) for task JSON, options, `pipeline_override`, interface imports, and locale requirements.
3. [references/troubleshooting.md](references/troubleshooting.md) for missing nodes, OCR/template mismatch, controller problems, and SceneManager misroutes.
4. [../../references/maa-framework-patterns.md](../../references/maa-framework-patterns.md) for verified MaaFramework binding and Pipeline semantics used across the repo.
5. Run `scripts/validate_pipeline_json.py` when you need a quick node-structure check, or `../../scripts/inspect_task_catalog.py` when checking task registry or option coverage.
## Operating Rules
- Coordinates, click targets, image templates, and ROI rectangles are based on 1280×720.
- Prefer nested v2 shape for new nodes:
```json
"recognition": {"type": "OCR", "param": {"roi": [0, 0, 100, 40], "expected": ["完整文本"]}},
"action": {"type": "Click"}
```
- Preserve existing local shorthand style when making a narrow edit in an old node, but do not mix incompatible forms inside the same field.
- Every action should follow recognition. A `DirectHit` is acceptable only after another node has proven state or for explicit config/entry placeholders.
- Do not solve missed states by adding broad loops or `max_hit` blindly. Add a state node, popup/loading handler, or clearer `next` branch.
- If a node truly needs no default wait, set `rate_limit`, `pre_delay`, and `post_delay` explicitly to `0`.
- Use `[JumpBack]` for popups, loading, scene transitions, and shared handlers that should return to the parent node.
- Use public SceneManager interface nodes from `Interface/Scene/`; do not route task pipelines directly to private `__ScenePrivate*` nodes.
- OCR `expected` values should use complete text; if a partial/regex string must skip i18n sync, mark that in source comments according to repo convention.
## New Task Checklist
1. Create or update `assets/resource/base/pipeline/<TaskFamily>.json` with a clear entry node.
2. Create or update `assets/resource/tasks/<TaskName>.json` with `task`, `entry`, `option`, and optional `controller` fields.
3. Add the task file to `assets/interface.json` imports.
4. Add or update five interface locale files when labels/descriptions use `$key`.
5. If Python is needed, coordinate with [../custom-actions/SKILL.md](../custom-actions/SKILL.md) for custom action registration.
6. Run safe checks: task catalog inspection, JSON formatting, `git diff --check`, and targeted Pipeline node/manual tests when a live environment exists.
## Common Review Questions
- Does the `entry` node exist and route from all realistic starting states?
- Does every `next` list cover success, loading, popup, wrong-scene, and completion states relevant to the task?
- Are controller restrictions accurate for the required input/screenshot mode?
- Are option names, input names, and `pipeline_override` target nodes exact?
- Are new OCR strings synchronized or intentionally skipped?
- Does the task avoid blind multi-click sequences and hidden implicit delays?
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!