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.
A no-code way to run Self-Consistency: have the AI solve the same problem via three different lines of reasoning, then compare the results and take the majority consensus. Especially useful for scenarios where a mistake is costly — tax estimates, loan interest comparisons, recipe scaling. Worth noting upfront: this uses more tokens, so it burns through free quotas faster.
A workflow mindset office workers can copy directly. Complex tasks (writing a proposal, building a report) done in one giant prompt come out poorly; this teaches you to split it into four steps — outline → expand each section → proofread and polish → convert to bullet points — with the output of each step feeding the next, plus four ready-to-use Traditional Chinese prompts. You can even change direction mid-task by rerunning just one step instead of starting over.
A reusable Traditional Chinese 'prompt optimizer' incantation — paste in a rough prompt and it first asks for the key missing details, then produces a well-structured version along with notes on what changed and why. Honestly flags its limits: a meta-prompt can fix structure and phrasing, but not factual errors in the content itself.
This doesn't write your resume — it audits one. It flags sentences that read like AI wrote them, uses pointed follow-up questions to force you to back up vague claims with real numbers, then gives a before-and-after rewrite example plus a personal banned-words list. Use it as the last check before you submit.
A turn-based mock interview: the interviewer asks scenario-based questions tailored to your industry, one at a time, follows up on your answer before giving feedback, and provides a speakable English version of the answer for each question. After eight questions, you get an overall hiring-likelihood assessment. Closer to a real interview than just reading past questions.
Nine times out of ten, asking AI for a formula fails because it doesn't know what your table actually looks like. This template gives you a fill-in-the-blanks format for describing your column structure; the formula you get back comes with a step-by-step breakdown, a 5-row mini sample table, and two common gotchas flagged, with both legacy and modern function syntax provided.
Generates the same email in three versions — formal, neutral, and friendly — each with a subject line, followed by a breakdown of "why it's worded this way." Especially useful for high-stakes emails like follow-ups, declines, apologies, and reporting up the chain — firm without burning the bridge.
Turn messy source material like meeting notes and report paragraphs into a slide outline formatted 'one point per page, separated by ---,' ready to paste straight into Gamma or Canva AI to produce a deck. Page titles are forced into statements of insight rather than labels, the conclusion goes on page two, and it comes with a 30-second opening script.
Use this right after any speech-to-text tool: paste the transcript and get a formal record, a list of decisions, an action-item table with "Owner/Deadline/Dependencies," and a 3-line summary for people who missed the meeting. It handles transcription typos and mixed Chinese-English text, flags any deadline that wasn't actually stated as "to be confirmed," and doesn't make things up.
Dump in a raw, whatever-you-jotted-down log of the week and get back a weekly report where results are quantified, blockers come with proposed solutions, and next week's plan is aligned to what your manager actually cares about. Scattered small tasks get merged automatically, slipping progress can't be hidden, and support requests are framed as needs — with an upward-management perspective built in.
For a boss's urgent demands, pushback, or last-minute extra work, this first diagnoses what they actually need (reassurance? a scapegoat? just an update?), then gives short, full, and escalation-ready reply options with risk notes for each. Built-in red-line check: it refuses goals that would require lying and offers an honest alternative instead.
Writing a ten-page proposal your manager doesn't have time to read is the real reason most proposals go nowhere. This prompt organizes your idea into a one-page proposal—problem, solution, expected benefit, resources needed, risks—all kept to a single page, and forces the AI to push back and ask you for missing information from a decision-maker's point of view. Good for situations where you need resources, budget, or headcount from a manager or another department.
Pulling a report but not knowing how to write the 'so what' is the most common sticking point in data work. This prompt has the AI act as a senior data analyst, interpreting the data you paste in through a three-layer structure—observation, hypothesized cause, recommended action—while forcing a hard distinction between 'conclusions directly supported by the data' and 'speculation that still needs verification,' so you don't get called out in a meeting for over-interpreting.
The biggest problem with handover documents is that everything you consider obvious never gets written down. This prompt has the AI act as a handover reviewer, first using structured questions to draw out the tacit knowledge in your head (who to contact, where passwords live, what has to happen every month, where the landmines are), then organizing it into a handover manual the successor can actually follow. Useful before resigning, transferring roles, or taking extended leave.
Installs a focused 'security auditor' subagent into Claude Code that reviews code and architecture file-by-file through an OWASP Top 10 lens as you develop features, classifying vulnerabilities by severity with directly-applicable remediation suggestions.
Installs a senior 'backend architect' subagent into Claude Code that uses contract-first design and domain-driven design to draw service boundaries, design REST/GraphQL/gRPC APIs, and build resilience and observability into the architecture from day one.
Installs a dedicated 'performance engineer' subagent into Claude Code that profiles code you've just written, finds bottlenecks like N+1 queries, memory leaks, and slow APIs, classifies them by impact, and provides before/after optimization recommendations with benchmark estimates.
A PostToolUse hook that makes Claude Code automatically run a formatter (Prettier/Black/gofmt) plus a linter every time it edits or writes a file—permanently removing 'inconsistent formatting' as a code review chore.
A PreToolUse Bash hook that intercepts destructive commands like rm -rf, fork bombs, and curl|sh before the AI executes them—a last line of defense that holds even against --dangerously-skip-permissions.
A Stop hook that fires a desktop notification and plays a sound the moment Claude Code finishes running. You can freely switch to other windows—the AI will call you back when it's done, no need to keep staring at the terminal.
A single npx command generates an information-rich terminal statusline for Claude Code, showing the current directory, Git branch, model, a context-usage progress bar, real-time spend burn rate, and a countdown to usage reset—hooked into live ccusage stats.
After Claude Code removed its built-in output style feature, this uses a SessionStart hook to automatically inject a custom response style (tone / format / rules) into every session, with a local marketplace for managing, activating, and applying multiple styles.
Brings obra/superpowers' "spec-first + true red-green TDD" development discipline into your agent: brainstorm to clarify requirements first, break the work into verifiable 2-5 minute tasks, then strictly follow RED-GREEN-REFACTOR, with a two-stage subagent review.
Borrowing the three-layer architecture from buildermethods/agent-os (Standards / Product / Specs): have AI first scan your codebase to extract existing conventions into documented standards, then auto-inject them every time it writes code — so you never have to re-explain the rules.