mattpocock-skills-zh-cn

This is the Simplified Chinese localized version of mattpocock/skills, providing a set of agent skills used by the author in daily real-world engineering work for coding agents such as Claude Code and Codex. It emphasizes small, modifiable, and composable skills rather than a methodology framework that takes over the entire development workflow. Localization motivation and strategy: - Chinese instructions reduce concept conversion costs for native Chinese-speaking developers; for models optimized for Chinese interaction or Chinese corpora, Chinese prompts better fit the Chinese context and reduce ambiguity from mixed Chinese-English content. - Translation adopts a skill-guided content localization approach: only natural language descriptions are translated, while directory names, skill names, frontmatter keys, commands, code blocks, paths, URLs, package/tool/API identifiers are preserved to avoid breaking installation and runtime behavior. - The repository syncs upstream by content refresh, without syncing upstream Git history or repository management state; the most recent translation sync was performed by Claude (Anthropic) and merged into main by the maintainer via PR. Maintenance rules are recorded in .skills/translate-skill/SKILL.md. Installation methods (choose one, do not install multiple times): 1. skills.sh installer: Run npx skills@latest add vinvcn/mattpocock-skills-zh-CN, then select the skills to install and target coding agents. This method copies skills into the project for easy modification and customization. 2. Claude Code plugin: Run /plugin marketplace add vinvcn/mattpocock-skills-zh-CN and /plugin install mattpocock-skills@mattpocock to install as a managed read-only bundle that updates uniformly with new versions, suitable for users who just want to use it directly and follow updates. 3. DeepSeek Harness (dsh): Install as a plugin with dsh plugin --profile web add @vinvcn/dsh-mattpocock-skills-zh, then invoke by prepending /zh- to the skill name, e.g. /zh-grilling. After first installation, select /setup-matt-pocock-skills and run it once in the agent to configure the issue tracker (GitHub, Linear, or local files), triage labels, and docs save location. Design philosophy: The author designed corresponding skills for common failure modes of coding agents: - Requirements misalignment: Through grilling sessions (/grill-me, /grill-with-docs), the agent asks detailed follow-up questions before starting work to align on intent. - Verbose expression: Shared language documents (CONTEXT.md) help the agent decode project terminology, making naming more consistent, code browsing easier, and thinking more token-efficient. - Code not running: Emphasizes feedback loops (static types, browser access, automated testing); /tdd drives the red-green-refactor cycle; /diagnosing-bugs provides a disciplined, stage-gated debugging loop. - Architecture decay: /to-spec asks which modules will be modified before writing a spec; /improve-codebase-architecture scans the codebase for deepening opportunities and generates a visual HTML report, positioned as a periodic survey rather than a rescue. Skills are categorized by who can invoke them into two types: user-invoked (requires the user to type the name to trigger, responsible for orchestration, can call model-invoked skills but not other user-invoked skills) and model-invoked (can be called by users or automatically triggered by the agent when a task matches, carrying reusable disciplines). Engineering: - User-invoked: ask-matt (skill router for this repo), grill-with-docs (grilling-style interview and maintains CONTEXT.md and ADRs), triage (advances issues via a triage roles state machine), improve-codebase-architecture, setup-matt-pocock-skills, to-spec (organizes conversation into a spec and publishes to the issue tracker), to-tickets (breaks a plan into tracer-bullet tickets with blocking relationships), wayfinder (maps large chunks of work into a shared map of decision tickets on the issue tracker), implement (implements per spec/ticket, drives /tdd at agreed seams, and concludes with /code-review). - Model-invoked: prototype (builds one-off prototypes, can produce single-file HTML or multiple UI variants), diagnosing-bugs, research (investigates against high-confidence primary sources and saves cited Markdown), tdd, domain-modeling, codebase-design (deep modules design discipline and vocabulary), code-review (dual-axis Standards and Spec review with parallel sub-agents), resolving-merge-conflicts (resolves conflicts hunk-by-hunk and traces intent from all sides, without using --abort), wizard (generates interactive bash wizards guiding manual completion of infrastructure configuration, credentials, CI secrets, one-off migrations, etc.). Productivity (general workflows, not limited to code): - User-invoked: grill-me, handoff (compresses conversation into a handoff document), teach (teaches across sessions using the current directory as a stateful teaching workspace), to-questionnaire (turns decisions into Markdown questionnaires for someone who can answer to fill out), wait-what (fills in missing context and restates the message in plain language). - Model-invoked: grilling (reusable interview primitive supporting grill-me, grill-with-docs, triage, wayfinder, improve-codebase-architecture), writing-for-agents (writes documentation for agents, such as AGENTS.md/CLAUDE.md). Misc (locally retained but rarely used): git-guardrails-claude-code (uses hooks to intercept dangerous git commands before push, reset --hard, clean, etc.), migrate-to-shoehorn (migrates as type assertions in test files to @total-typescript/shoehorn), scaffold-exercises (generates exercise directories with sections, problems, solutions, explainers), setup-pre-commit (configures Husky pre-commit hooks integrating lint-staged, Prettier, type checking, and tests).