LingquLab Skills

LingquLab Skills is a Git-backed ZCode and Codex marketplace for independently installable skills and skill series.

Install the Marketplace

In ZCode, open Settings → Plugins → Create → Add marketplace and enter:

LingquLab/harness

The repository-root marketplace.json is the ZCode catalog. Each plugin provides a preferred .zcode-plugin/plugin.json manifest and a compatible Codex manifest.

For Codex, add the same repository from the command line:

Add this repository as a Codex marketplace:

codex plugin marketplace add LingquLab/harness

Then install a plugin from the catalog:

codex plugin add superpowers-neo@lingqulab

Start a new Codex task after installation so newly installed skills are discovered reliably.

Plugins

Plugin	Description	Version
superpowers-neo	Pragmatic software-development workflows with rigor scaled to task complexity and risk	0.3.3
ascendc-development	Version-aware Ascend C operator development, API guidance, review, diagnostics, and CANN setup workflows	0.3.1
cross-zone-development	Secure GitHub issue handoffs between blue development and green service-debugging zones	0.1.4
persistent-shell	Reusable stateful SSH shell sessions from Windows	0.1.1
archify	Validated, interactive architecture and system diagrams	2.17.0-dev.1

Archify

Install the plugin from the marketplace:

codex plugin add archify@lingqulab

Its archify skill turns system descriptions or repository evidence into validated architecture, workflow, sequence, data-flow, and lifecycle diagrams. The vendored runtime produces self-contained HTML and supports image and WebM export. It is pinned to upstream commit 18911058008f17dc065af23a2cdc9bfeff6d3f7a; see the plugin's vendor record and third-party notices.

Persistent Shell

Install the plugin from the marketplace:

codex plugin add persistent-shell@lingqulab

Its persistent-shell skill installs a shared PowerShell and Git Bash pshell launcher backed by a user-level local daemon. Repeated remote commands reuse one non-interactive SSH shell, preserving state such as the working directory and environment variables. The installer discovers an available Python 3 interpreter with Paramiko instead of requiring a fixed Python minor version.

Cross-Zone Development

Install the plugin from the marketplace:

codex plugin add cross-zone-development@lingqulab

Its single cross-zone-development skill follows a direct blue- or green-zone designation from the user regardless of model family. Without one, GPT/OpenAI defaults to blue and DeepSeek/GLM defaults to green; unknown identities ask the user before acting. Green may modify and validate code locally, but it returns only concrete, bounded diagnostics—not code, changes, bulk logs, internal endpoints, sensitive payloads, or exported artifacts.

Ascend C Development

Install the plugin from the marketplace:

codex plugin add ascendc-development@lingqulab

The plugin contains eight independently triggered skills:

Skill	Use when
ascendc-api-best-practices	Implementing or debugging Ascend C API usage and alignment, buffer, precision, or pipeline behavior
ascendc-code-review	Reviewing Host, Tiling, Kernel, SIMT, build, or operator changes
ascendc-docs-search	Locating version-matched local or official Ascend C documentation and examples
ascendc-env-check	Performing read-only CANN environment and NPU visibility diagnostics
ascendc-npu-status	Checking JSON-listed Ascend NPU process occupancy locally or over SSH
ascendc-operator-development	Developing or migrating a complete registered or direct-launch Ascend C operator through layered validation
ascendc-runtime-debug	Diagnosing runtime, Tiling, launch, device-exception, hang, precision, or performance failures from bounded evidence
cann-env-setup	Planning or carrying out a guarded, version-matched CANN installation or repair

The original five skills are adapted from TileXR's Claude skills at source commit 1e2619e793b5894a1aec2d7d6897dbe5f7c501c0. The operator-development and runtime-debugging workflows extend that baseline without importing cannBot automation. Claude-specific tool calls, fixed environment assumptions, destructive diagnostic commands, and the duplicate commit-push-pr skill are intentionally not shipped. Useful online workflows remain scripted: one dependency-free client searches and fetches Huawei's official documentation, and another obtains public GitCode PR diffs through bounded shallow fetches. See the migration audit, expansion plan, and third-party notices.

Superpowers Neo

Superpowers Neo is a modular software-development workflow for coding agents. It keeps explicit design, plan execution, debugging, review, verification, and Git delivery practices while scaling ceremony to the ambiguity and risk of the task.

Neo has no global entry skill. Each skill is independently discoverable and loads only when its own trigger matches.

Skill	Use when
superpowers-neo-designing-complex-changes	A change is complex, ambiguous, cross-component, or architecture-sensitive
superpowers-neo-writing-plans	Approved work needs a multi-step or subagent-ready implementation plan
superpowers-neo-using-git-worktrees	Workspace isolation may be needed for dirty or parallel work
superpowers-neo-executing-plans	An in-scope plan is ready for main-agent and scoped-subagent execution
superpowers-neo-validation-strategy	A change needs validation proportional to its risk
superpowers-neo-systematic-debugging	A bug or unexpected failure needs evidence-based diagnosis
superpowers-neo-code-simplification	Implemented code should be simplified without changing behavior before final review or delivery
superpowers-neo-requesting-code-review	A substantial or risky change benefits from independent review
superpowers-neo-handling-code-review-feedback	Review feedback needs technical evaluation
superpowers-neo-verification-before-completion	Work is about to be described as complete, fixed, or passing
superpowers-neo-git-delivery	Completed Git work needs scoped commit and delivery handling

What Changes from Superpowers

• No using-superpowers startup or umbrella skill.
• Complex-change design and persistent plans trigger only when complexity justifies them.
• Worktrees and subagents are selected by isolation and coordination value.
• Validation is risk-driven; test-first development is useful but not universal.
• Completed task code is simplified before final validation and review; a no-op is valid when the code is already clear.
• Independent review is selected by risk rather than required after every task.
• Review findings require a concrete reachable failure and material impact; speculative defensive handling and warn-and-continue fallbacks are not improvements by default.
• Automatic delivery authorizes scoped task commits and normal pushes from established task-owned non-default branches. Manually invoking superpowers-neo-git-delivery additionally authorizes task-branch creation, normal push, and PR creation; merge, history rewrite, force push, hook bypass, and cleanup remain separately protected.
• Skill-authoring methodology is not part of the shipped series.

The code-simplification workflow is primarily adapted from caarlos0/dotfiles/skills/code-simplifier at source commit b2c38ba14c4295476f4672bb097a405edd992642. Keep this pinned source when checking for upstream improvements; Neo intentionally retains only the general, behavior-preserving workflow rather than copying its language-specific guidance.

See the Superpowers Neo design for its behavior contract and the marketplace design for packaging and extension decisions.

Validate

The repository validator uses Ruby standard libraries and needs no package installation:

ruby scripts/validate-skills.rb
ruby -c scripts/validate-skills.rb
python3 scripts/validate-zcode.py
bash -n scripts/install.sh

These checks cover both marketplace catalogs, all Codex and ZCode manifests, localized plugin documentation, skill packages, relative documentation links, and plugin behavior scenarios.

Behavioral validation is a fresh-agent evaluation. Give a new agent only the relevant SKILL.md files and the request section from one file under tests/<plugin-name>/scenarios/, then compare the response with its expected behaviors and failure signals. Do not include the expected result in the agent prompt.

Before publishing a plugin change, also run the current validator supplied by Codex's installed plugin-creator skill against that plugin directory. This catches schema changes that may be newer than the repository validator.

Manual Superpowers Neo Installation

Marketplace installation is preferred. A manual fallback remains available for environments that do not use Codex plugins.

Preview the installation:

scripts/install.sh --dry-run

Install to ${CODEX_HOME:-$HOME/.codex}/skills:

scripts/install.sh

Use --target PATH to install elsewhere. The installer copies exactly the eleven Neo skill directories from plugins/superpowers-neo/skills/, refuses to overwrite existing targets, and never disables or removes the original Superpowers plugin.

Avoid Duplicate Installations

Manual copies under ${CODEX_HOME:-$HOME/.codex}/skills/superpowers-neo-* can coexist with the marketplace plugin, but duplicate skill names make the active source ambiguous. After validating the marketplace plugin in a new Codex task, deliberately remove or archive the old manual copies. Marketplace installation does not modify them automatically.

The original superpowers@openai-api-curated plugin may also coexist with Neo during evaluation. Remove it only after Neo has been reviewed in real tasks and the user explicitly authorizes that cutover.

Add a Plugin

Each independently versioned plugin lives under:

plugins/<plugin-name>/
|-- .codex-plugin/plugin.json
|-- .zcode-plugin/plugin.json
|-- README.md
|-- README_CN.md
`-- skills/

To add a plugin:

1. Use one normalized lower-case hyphenated name for its directory, manifest, and marketplace entry.
2. Give the plugin its own strict semantic version.
3. Put runtime skills under plugins/<plugin-name>/skills/.
4. Register it in root marketplace.json for ZCode and .agents/plugins/marketplace.json for Codex; catalog order is user-visible.
5. Keep the plugin version synchronized across both manifests, the root marketplace entry, and this README.
6. Use a supported lower-case ZCode category and include the Codex installation and authentication policy fields.
7. Provide semantically equivalent README.md and README_CN.md files covering invocation, dependencies, network access, commands, file effects, Hooks, MCP, provenance, and licensing.
8. Add plugin-specific tests under tests/<plugin-name>/ and extend repository validation.
9. Document its Codex selector as <plugin-name>@lingqulab.

Keep coherent skill series together, but publish unrelated skills as separate plugins instead of expanding one catch-all package.

Attribution

Superpowers Neo is an independent adaptation inspired by Superpowers by Jesse Vincent. Superpowers is available under the MIT License. See THIRD_PARTY_NOTICES.md for the upstream notice. This project is not affiliated with or endorsed by the upstream project.

License

This marketplace and Superpowers Neo are licensed under the MIT License. The ascendc-development plugin is separately licensed under the CANN Open Software License Agreement Version 2.0; see third-party notices.