ZHENGGUANG LI
Community@zhengguangli
ZHENGGUANG LI maintains a 13-skill harness system for Claude Code that governs agent-driven project intake, planning, verification, and commit quality gates.
Agent Skills by ZHENGGUANG LI
Showing 9 vetted skills indexed across 1 GitHub repositories.
zhengguangliharness-project-intake
Analyzes project repositories and generates structured project cards covering tech stack and architecture.
zhengguangliharness-repo-map
Restructures bloated CLAUDE.md files into a map plus structured docs/ knowledge base.
zhengguangliharness-authoring
Guides authoring of SKILL.md files and subagents with context budget discipline.
zhengguangliharness-exec-plans
Persist complex multi-session tasks as versioned execution plans with verifiable steps and decision logs.
zhengguangliharness-architecture-boundaries
Designs and audits layered architecture dependency rules via read-only Grep and Bash checks.
zhengguangliharness-bootstrap
Generates CLAUDE.md maps, docs/ skeletons, and .gitignore rules to initialize project harness structures.
zhengguangliharness-golden-principles
Encode code review feedback as lint rules and run periodic scans generating small fix PRs.
zhengguangliharness-observability-and-browser
Verify code changes using browser automation and observability signals instead of static inference.
zhengguangliharness-orchestration
Route user goals to skill combinations and execution order across five standard workflows.
Frequently Asked Questions About ZHENGGUANG LI
FAQPage SchemaWhat tasks can I accomplish with ZHENGGUANG LI's harness skills?▼
You can analyze project tech stacks, slim bloated CLAUDE.md files, optimize LLM prompts, persist multi-session execution plans, enforce layered architecture boundaries, run pre-commit quality gates, and verify UI changes through browser automation and observability signals.
Who should use these harness skills?▼
Engineers using Claude Code who manage agent-generated code at scale. It suits developers needing structured project knowledge, teams fighting architecture drift and circular dependencies, and anyone wanting auditable verification records before merging code changes.
How do the harness skills run in practice?▼
Each skill runs in a forked context within Claude Code, triggered explicitly by user phrases or implicitly by project state. They operate through restricted Bash commands like git, grep, rg, npm, vitest, and tsc, with some skills writing files via Write and Edit permissions.
What are the dependencies between the harness skills?▼
harness-project-intake is the foundation for repo-map, prompt-optimizer, architecture-boundaries, bootstrap, and golden-principles. The commit gate depends on verification-loop, observability, and exec-plans, forming a layered chain from intake through planning to verified commits.
What prerequisites are needed to use these skills?▼
You need a Claude Code environment, since every skill declares claude-code compatibility. Verification and commit-gate skills additionally require project-level test, build, or lint configurations such as npm, bun, cargo, vitest, or make to provide feedback signals.