1328 hand-tested AI prompts across image, video, and music — each with full context, variables, and a sample output. Content is authored in Traditional Chinese; the UI here is English.
Curates three core project-rules from steipete/agent-rules into a single rule pack usable directly as a slash command: semantic commits (12 types plus emoji, with automatic splitting into atomic commits), a 6-persona professional PR review, and 5 Whys root-cause analysis.
Borrows the approach from the 40k-star awesome-cursorrules repo to codify 'how to write for a given tech stack' into a project rule file, so Cursor / Claude Code / Codex always produce code in your conventions instead of generic boilerplate.
Has Claude Code scan your repo and produce a tight, verifiable CLAUDE.md project memory file under 200 lines—covering build/test/lint commands, directory structure, conventions, and prohibitions—while routing anything that doesn't belong there into skills or path-scoped rules instead.
Turns the checklist or multi-step process you re-paste every time into a custom slash command (.claude/commands/<name>.md, or the newer .claude/skills), with correct frontmatter, $ARGUMENTS parameters, live context pre-injected via `!cmd` (git diff, PR content), and permissions scoped down with allowed-tools.
Uses hooks in .claude/settings.json to automatically run formatting, linting, type-checking, or tests at specific Claude Code lifecycle events (PostToolUse after an edit, PreToolUse before running a command, Stop when wrapping up)—producing correct JSON config plus an executable hook script, with exit code 2 hard-blocking dangerous actions.
Has Claude Code interview you about a repetitive manual process, then produce a fully-structured Agent Skill (SKILL.md) covering name/description/trigger conditions/step-by-step instructions/acceptance criteria/supporting references, and verify it can be triggered automatically as well as invoked via a /slash command.
Has Claude Code help you connect an MCP server (a database, Notion, GitHub, your own API...), choosing the right transport and scope, using .mcp.json or claude mcp add, handling secrets and OAuth securely, and finally demonstrating how mcp__server__tool tools and @resource pull external data into your workflow.
Has Claude Code break a large task into several independent, parallel subagents (each with its own context, restricted tools, and an assignable model), planning how to fan the work out, defining each subagent's responsibility boundaries and reporting format, and producing reusable .claude/agents subagent definition files when needed.
Applies the spirit of Plan Mode so Claude Code hands over a structured plan (files affected, steps, risks, acceptance criteria) before touching any file—you review, correct, and approve it before execution begins. Prevents the AI from charging ahead and writing a pile of things you didn't want.
Forces Claude Code through a strict red→green→refactor cycle: for every behavior, first write a test that fails (red) and actually run it to prove the failure, then write the minimal implementation that just makes it pass (green), and finally refactor under the safety of the passing tests. Cures the AI's old habit of jumping straight to the happy path, skipping edge cases, and faking test coverage.
Corrects Claude Code's tendency to 'see an error and start randomly editing' into a disciplined 4-step debugging process: reliably reproduce the bug first, isolate the true root cause (not just the symptom) through minimization, write a regression test that fails, apply the smallest possible fix, then run the full test suite to prove it's fixed without breaking anything else.
Has Codex do a disciplined code review of your git diff across three tracks (correctness bugs / simplification-and-reuse / security), listing each finding with a confidence level and a pastable fix, and explicitly instructed to say so when nothing's wrong instead of manufacturing complaints.
Forces Codex through a closed loop — write a failing test that reproduces the bug, find the root cause, make the minimal fix, watch that same test go from red to green, then run the full test suite to confirm no regressions — eliminating both 'fixed it without verifying' and 'changed a bunch of unrelated stuff along the way.'
Has Codex write characterization tests for untested legacy code: it first reads and documents what the code 'actually does right now' (bugs included), systematically covers the happy path plus edge cases and error paths, and is strictly forbidden from modifying the code under test.
Has the agent automatically inventory your Cursor project (.cursorrules, .cursor/rules, .cursor/mcp.json), produce a step-by-step migration plan, and generate the corresponding CLAUDE.md or AGENTS.md, slash commands, and MCP config — so you're not stuck 'dumber than Cursor' on day one.
A Definition of Done template you can paste into any coding agent, turning 'what counts as done' into a checklist the agent must self-verify item by item: behavior meets acceptance criteria, tests/lint/typecheck/build all pass (or it states why they didn't run), risky changes come with a rollback plan, and a final 'verification story' is delivered. Puts an end to agents claiming 'done' while everything is actually broken.
A large-refactor playbook for agents: first do a read-only survey of the 'blast radius' — which files, call sites, and tests are affected — then split the change into small batches, each preserving behavior, each run through its own tests, each its own atomic commit. Puts an end to 'change 40 files at once, then the build breaks and no one knows where to start fixing it.'