cc-switch-web: Skill for Claude Code

.pi/skills/xirang-build/SKILL.md

xirang-build is a skill for Claude Code, Codex from cp-yu/cc-switch-web. It costs 20 tokens per session (2,109 once invoked), scanned A, original, MIT.

A skill for building or rebuilding a project's Xirang Semantic Model as one reviewed candidate. The model is a structured record of the project's intended parts, rules, and relationships.

In plain words
What is it for?
It is for establishing or reconstructing the model that describes the project's elements, relationships, and views.
Why use it?
It creates a reviewable representation of project intent from authorized instructions and available evidence.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions subagents.

This is cp-yu/cc-switch-web's own configuration. It tells Claude Code and Codex how to work on cc-switch-web itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything cc-switch-web configures →

Reuse

Borrowing it

Nothing to install: this file belongs to cp-yu/cc-switch-web. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/cp-yu/cc-switch-web/main/.pi/skills/xirang-build/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/cp-yu/cc-switch-web

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for xirang-build

README.md
[![agentmods](https://agentmods.dev/badge/skills/cp-yu/cc-switch-web/xirang-build/github.svg)](https://agentmods.dev/skills/cp-yu/cc-switch-web/xirang-build)
Your own site
<a href="https://agentmods.dev/skills/cp-yu/cc-switch-web/xirang-build"><img src="https://agentmods.dev/badge/skills/cp-yu/cc-switch-web/xirang-build/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for xirang-build

Your own site · 80×15
<a href="https://agentmods.dev/skills/cp-yu/cc-switch-web/xirang-build"><img src="https://agentmods.dev/badge/skills/cp-yu/cc-switch-web/xirang-build.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 20 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,109 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00020 $0.02109
Opus 5.5 $0.00008 $0.00844
Sonnet 5.5 $0.00004 $0.00422
Haiku 4.5 $0.00002 $0.00211

Measured 1mo ago against content hash b95ebb996a3f, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

xirang-build scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.pi/skills/xirang-build/SKILL.md · 91 lines

How it starts

The opening of the file, as written. The whole thing — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Build the project Xirang Semantic Model from user-authorized intent and evidence.

Xirang Philosophy

  1. Xirang is a structured representation of human intent that an Agent can compile.
  2. One Xirang Semantic Model is persisted as a single whole in the four partitions metamodel/, elements/, relationships/, and views/; a Semantic Delta uses the same four partitions and adds operation.
  3. A change reconciles a Semantic Delta toward the target steady state. proposal.md, design.md, and tasks.md are compilation scaffolding, not competing sources of truth.
  4. The Xirang Semantic Model is complete only when an Agent need not guess decisions that affect element hierarchy, contracts, or relationships.
  5. The Agent acts like a compiler and faithfully translates authorized human intent. Existing code is current implementation evidence and MUST NOT silently override the Xirang Semantic Model.

Element Contract Semantics

  • Element Contract SHALL 完整表达宿主 Element 在自身抽象层级承担的职责、保证、约束与行为;children 可以进一步精化或共同实现这些承诺,父子 Elements 可以在各自层级表达相互覆盖的完整语义。
  • Requirement SHALL 以稳定 identity 表达一项可独立演进的规范承诺。以该承诺能否独立新增、修改或移除判断边界,不得按句子、分句、SHALL 数量或目标条数机械拆分。只复述 Declaration summary 或 sibling Requirements 语义并集且不增加规范承诺的内容不形成 Requirement;独立的不变量、顺序、原子性、一致性或完成条件应保留。
  • Scenario SHALL 是具有规范约束力的 Requirement 组成,只具体化宿主 Requirement 在特定条件下的行为,不得引入可独立演进的承诺。Scenarios 不默认穷尽 Requirement 的全部适用情况,Scenario 不作为独立 Semantic Delta Entry,其变化由宿主 Requirement 的完整目标内容表达。

Semantic Model Unit Notation

Every Markdown unit declares its own entity in frontmatter. The partition does not determine the type.

entity frontmatter fields
element-declaration identity, kind, parent, title, summary
element-kind identity, contract; optional root, parents, children
relationship-kind identity; optional sourceKinds, targetKinds
authored-view identity, include; optional of, title, autoLayout
  • parent: null marks the single Project Root. include is '*' or a list of element identities; of is one element identity.
  • A relationships/ file is a container of {source, kind, target} entries. A Relationship's identity is its entire content and it carries no other field.
  • An elements/ unit body is exactly the ## Requirements section: ### Requirement: <name> with #### Scenario: <name> beneath it. Descriptive prose belongs to the Declaration's summary and MUST NOT be repeated in the Contract. Any other body content is a validation error.
  • A views/ unit has no body.
  • identity uses [A-Za-z0-9._-]+ and contains no path separator. Element Kind and Relationship Kind identities are globally unique within the Metamodel. Do not encode a parent path or hierarchy position into an identity: position changes over time while identity does not.
  • Default file naming, non-normative: elements/<identity>.md, metamodel/<kind identity>.md, views/<view identity>.md, relationships/<relationship kind identity>.yaml grouped by Relationship Kind.

Workflow

  1. Confirm that the project has been prepared with xirang setup, then run xirang candidate status --json.
  2. If a Candidate is active, present its baseline, inventory, and status. Ask the user to choose either to continue the active Candidate or explicitly authorize discarding it and initialize a replacement. Build MUST NOT silently continue, discard, or replace an active Candidate.
  3. Ask the user to choose the exploration scope: whole project, code and tests, documentation and current Xirang, or custom paths and rules.
  4. Resolve the Candidate baseline before authoring any Build file. Summarize the Formal Model when it exists, ask the user to choose the starting point, then run exactly one applicable command:
    • xirang candidate init --from current
    • xirang candidate init --from clean
    • xirang candidate init --from-path <path>
  5. After initialization succeeds, write .xirang/candidate/build.md with the authorized scope, authority order, current requirements, and explicit exclusions. Current user requirements have highest priority.
  6. Explore the authorized scope in any useful order. Code, tests, documents, configuration, Git history, and current Xirang are evidence only unless the user explicitly designates them as source of truth. Subagents MAY accelerate read-only exploration.
  7. Run the conditional Modeling Decision Gate after exploration and before the first Candidate model write.
    • Ask only when multiple reasonable choices would change identity, hierarchy, Contract, Kind, Relationship, or Authored View semantics. Do not block on implementation details or wording preferences that cannot change the target Semantic Model.
    • Resolve decisions in dependency order: authority conflicts → Element identity and boundaries → hierarchy → Metamodel → Contracts → Relationships → Authored Views.
    • Present bounded options with trade-offs and ask one decision at a time. If no semantic choice remains, continue without asking.
    • Append only exception provenance to build.md: user rulings, authority-conflict resolutions, non-obvious evidence choices, and explicit exclusions. Record each decision, evidence, and affected scope; do not create a Requirement provenance matrix.
  8. Author the Candidate breadth-first across the whole model: Metamodel → Element Declarations and hierarchy → Element Contracts → Relationships → Authored Views.
    • Write .xirang/candidate/{metamodel,elements,relationships,views}/ in the unit notation above. An Element Contract is the body of its Element unit, so one Element has at most one Contract; whether a Contract is required comes from the contract field of its Element Kind.
    • Check each completed layer with focused Agent inspection. Do not require full candidate validate for an intentionally incomplete intermediate layer.
    • If a later layer exposes an earlier defect, correct the affected layer and recheck every dependent later layer without rewriting unrelated layers.
  9. Run xirang candidate validate --json after all five layers exist. Fix every ERROR in Candidate source and repeat deterministic validation until it succeeds. The CLI is read-only and must not author or normalize semantics.
  10. After deterministic validation succeeds, delegate one complete semantic review to a generic read-only subagent with a clean context.
    • Provide the absolute project root, Candidate root, build.md, current reviewDigest, authorized scope, authority order, recorded user rulings, validation result, review checklist, and output contract.
    • Require the subagent to independently read build.md, all four Candidate partitions, authority sources, and necessary project evidence. Main-Agent completion claims are not evidence.
    • Review authorization coverage, unauthorized durable semantics, hierarchy, Kinds, Contracts, Requirement boundaries, Scenario confinement, Relationships, Authored Views, cross-level overlap, and exception provenance.
    • Only BLOCKER and HIGH findings fail the gate. Correct a finding or return to the Modeling Decision Gate as appropriate.
    • Any Candidate modification invalidates the review. Rerun xirang candidate validate --json and delegate another new clean-context subagent; never resume or reuse the previous review context.
    • If a clean-context subagent is unavailable, fail closed. Do not substitute author self-review or use xirang-reviewer.
  11. Present the Project Root, Metamodel, hierarchy, Element Contracts, Relationships, Authored Views, important confirmed decisions, Formal comparison, semantic-review result, and returned reviewDigest directly to the user.
  12. Only after the user confirms that exact version, run xirang candidate promote --digest <reviewDigest>. Promotion replaces .xirang/model/ with the Candidate as a whole; a unit absent from the Candidate is not retained.
  13. After promotion succeeds, verify these postconditions in order:
    • Run xirang candidate status --json and require active === false.
    • Verify .xirang/model/metamodel, .xirang/model/elements, .xirang/model/relationships, and .xirang/model/views are real directories, even when a partition is empty.
    • Run xirang arch validate --json and require success. Report warnings without failing the postcondition.
    • Report Build complete only after all checks pass. On failure, report whether promotion succeeded, the failed condition, and current Formal/Candidate state; MUST NOT retry promotion automatically.

Read the full file on GitHub · 91 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 1mo ago First seen · 91 lines · 20 tokens per session scan A b95ebb996a3f

Subscribe to this mod's changes

xirang-build is a skill published in the GitHub repository cp-yu/cc-switch-web (47 stars, last pushed 1mo ago), licensed MIT. It adds 20 tokens to every session and 2,109 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-12.

Related

Other skills, from other repositories

local-ai-agents

Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…

microsoft/ai-agents-for-beginners · 200 tokens

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.

obra/superpowers · 21 tokens

next-cache-components-adoption

Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…

vercel/next.js · 95 tokens

chat-pet-sprite-creation

Use when creating or changing VS Code chat pet sprite art, sprite sheets, state animations, eye treatments, Stable/Insiders variants, or pet transitions under src/vs/workbench/contrib/chat/browser/widget/media/chatPet.

microsoft/vscode · 53 tokens

cpu-profile-analysis

Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…

microsoft/vscode · 71 tokens

insight-error-page

Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…

vercel/next.js · 83 tokens