Install
$ agentstack add skill-twlongzai-agent-skill-write-like-human ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Write Like Human
Core Purpose
Write prose that feels authored by a person with a point of view, not generated from a template. Keep the author's intent, facts, constraints, and voice in control; use AI only to accelerate structure, drafting, and editing.
Prefer clear, specific, grounded writing over ornate, inspirational, or overexplained writing.
Operating Principles
- Put the writer before the tool.
- Preserve the user's ideas, experience, claims, uncertainty, and vocabulary.
- If the user supplies a draft, treat it as the source of truth and revise from it.
- If the user supplies voice samples, infer a compact style guide before rewriting.
- Make every sentence earn its place.
- Keep only sentences that add information, sharpen meaning, move the argument, or create necessary rhythm.
- Remove repeated explanations, synonym loops, and "in other words" restatements unless the audience truly needs them.
- Choose concrete detail over emotional labeling.
- Show emotion through situation, choice, image, sequence, or contrast.
- Avoid blunt labels like "lonely," "heartbreaking," or "deeply meaningful" unless the directness fits the genre.
- Keep the scale honest.
- Do not turn small observations into grand moral conclusions.
- Let modest things stay modest. Let important things prove their importance through evidence.
- Use AI as an editor, not a ghost identity.
- Do not fabricate personal history, metrics, citations, experience, sources, or feelings.
- For resumes, applications, academic work, and policy-bound submissions, respect the user's institution or employer rules. If AI-written content is prohibited, offer coaching, outline help, critique, or revision suggestions instead of producing deceptive final copy.
Workflow
1. Identify the Job
Before writing or revising, infer or ask for only the missing details that materially affect the result:
- Purpose: inform, persuade, reflect, apply, sell, explain, apologize, announce, teach.
- Audience: reader's knowledge level, relationship to writer, stakes, attention span.
- Format: article, essay, email, resume bullet, cover letter, report, post, script, memo.
- Voice: plain, warm, sharp, reflective, professional, casual, expert, restrained.
- Constraints: length, language, required facts, banned topics, citation needs, disclosure rules.
When the user gives enough context, proceed without interrogation.
2. Build From the User's Material
Use the user's facts and phrasing first. When creating from scratch, ask for or clearly mark missing lived detail instead of inventing it.
For personal or professional writing, seek the "proof layer":
- What happened?
- Who was involved?
- What changed?
- What did the writer do?
- What specific result, metric, artifact, constraint, or lesson can be verified?
- What detail would only this writer know?
For resumes and cover letters, start from the user's own draft or notes, then improve fit, clarity, and evidence. Avoid overpolished claims that the user cannot defend in an interview. Prefer concrete context and CAR-style evidence: context, action, result.
3. Draft Simply First
Write a clean version before adding style. Use ordinary verbs and direct syntax.
Prefer:
- "decide" over "make a decision"
- "discuss" over "conduct a discussion"
- "test" over "perform testing"
- "explain" over "provide an explanation"
- "use" over "leverage" unless the business context truly calls for it
Aim slightly simpler than the requested register when the task tends to produce pompous prose. This is especially useful for school-style reflections, general essays, and personal statements: a simpler baseline often sounds more human than a falsely mature one.
4. Add Human Specificity
Add specificity only when it is true or clearly supplied:
- One concrete scene instead of three abstract claims.
- One precise example instead of a broad compliment.
- One measured result instead of vague excellence.
- One honest uncertainty instead of fake confidence.
- One lived observation instead of a generic lesson.
Use first person when the genre permits it and when it carries real judgment or experience. Do not add first person as decoration.
5. Edit for Rhythm
Vary sentence length. Let some sentences be short.
Use longer sentences for context, qualification, and causal links. Use short sentences for emphasis, transition, or restraint.
Read the draft aloud mentally. Fix places where:
- clauses stack without control,
- every sentence has the same shape,
- the paragraph sounds like a presentation script,
- the ending lands with a slogan instead of a thought.
6. Cut the AI Smell
Run the anti-pattern checklist before finalizing:
- False antithesis: Avoid "not X, but Y" unless both sides are genuinely opposed and the contrast is necessary.
- Repetitive paraphrase: Do not say the same point three ways.
- Forced emotion: Do not explain what the reader should feel.
- Forced elevation: Do not inflate small events into universal truths.
- Forced drama: Remove manufactured transitions like "but the real problem was only beginning."
- Empty attribution: Avoid "some people say" unless you can identify who says it or why the claim matters.
- Generic empathy: Avoid therapy-like validation such as "you are not alone" unless the genre calls for it.
- Fake sincerity: Avoid stock openings like "honestly," "the truth is," or "what most people miss is" when they add no information.
- Over-signposting: Cut "in this article I will explain," "let's dive in," and "here is a detailed breakdown" when the structure is already clear.
- Over-formatting: Do not rely on excessive bold text, emojis, one-line paragraphs, or decorative headings to create energy.
- Jargon padding: Replace vague prestige phrases with the actual action, claim, or evidence.
- Nominalization: Prefer verbs over noun-heavy constructions.
- Translation residue: Remove unnecessary pronouns and filler transitions created by English-like sentence patterns in non-English prose.
7. Preserve the Voice
When revising a user's draft, create a quick internal style guide:
- sentence length and rhythm,
- preferred vocabulary,
- level of formality,
- use of humor, doubt, or directness,
- typical paragraph shape,
- recurring metaphors or avoided phrases.
Make the smallest changes that solve the problem. If the user asks for flow, fix flow. If the user asks for grammar, do not overwrite their personality. If the user asks for stronger writing, improve structure, verbs, and evidence before changing tone.
8. Structure Without Sounding Mechanical
Use structure to help the reader, not to show that the text is structured.
For workplace writing:
- Put the main point early.
- Use enough context for decisions.
- Separate facts, interpretation, and asks.
- Prefer concise paragraphs and clear next steps.
For essays and articles:
- Start from a real tension, question, observation, or scene.
- Let the thesis emerge cleanly.
- Use examples before broad claims when possible.
- End by landing the specific point, not by forcing a universal moral.
For educational or reflective writing:
- Keep the student's or author's likely vocabulary.
- Avoid over-mature phrasing.
- Use direct observations and modest conclusions.
- Do not over-explain obvious lessons.
For brand and marketing writing:
- Require brand voice, audience, offer, differentiation, and proof points.
- Avoid copy that could apply to any competitor.
- Watch for copyright and originality risks when adapting examples.
- Use AI to generate options, then select and rewrite around the real brand position.
9. Verify and Own the Draft
Before final output, check:
- Facts are accurate or clearly framed as assumptions.
- Claims can be defended.
- The text matches the requested audience and format.
- The strongest original details survived the edit.
- No sentence exists only to sound profound.
- No paragraph repeats the previous paragraph.
- The final line sounds like a natural close, not a quote poster.
If shortening, preserve the user's strongest lines and core ideas first. Do not compress by deleting the parts that carry voice, specificity, or proof.
Response Patterns
When drafting:
- Produce the requested text directly.
- Keep style natural and specific.
- Add a short note only if assumptions, missing facts, or risk matter.
When revising:
- Preserve meaning and voice.
- Improve clarity, rhythm, specificity, and evidence.
- Avoid explaining every edit unless the user asks.
When the source text is weak:
- Strengthen structure and verbs first.
- Ask for concrete details if the writing needs lived evidence.
- Mark placeholders plainly rather than inventing details.
When the request risks dishonesty:
- Refuse deceptive ghostwriting or fabricated credentials.
- Offer an acceptable alternative: outline, critique, practice version, editing checklist, or coaching questions.
Quick Final Pass
Use this compact final pass before sending prose:
- Cut one generic sentence.
- Replace one noun-heavy phrase with a verb.
- Add or preserve one concrete detail.
- Vary one sentence rhythm.
- Remove one forced conclusion.
- Make sure the piece sounds like someone meant it.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: twlongzai
- Source: twlongzai/agent-skill
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.