# Write For Humans > >- - **Type:** Skill - **Install:** `agentstack add skill-tschuehly-skill-incubator-write-for-humans` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [tschuehly](https://agentstack.voostack.com/s/tschuehly) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [tschuehly](https://github.com/tschuehly) - **Source:** https://github.com/tschuehly/skill-incubator/tree/master/skills/write-for-humans ## Install ```sh agentstack add skill-tschuehly-skill-incubator-write-for-humans ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Write for humans Minimize the reader's effort to find the point, understand it, and act on it. Make the text: - **relevant:** it serves the reader's goal; - **findable:** its answer, evidence, and actions are easy to locate; - **understandable:** its language and relationships are clear to the intended reader; - **usable:** the reader can decide, act, or continue without reconstructing missing context. ## 1. Model the reader Before drafting, infer from the request and available context: - what the reader wants to understand, decide, or do; - what the reader probably already knows; - which facts, terms, and constraints are new to them; - how much detail they need now. Match the language to the reader's knowledge rather than to superficial vocabulary. Preserve expert depth for expert readers. If a missing fact would materially change the text, ask one short question. Otherwise, choose a reasonable reader model and proceed. This step is complete when the reader's goal, knowledge, needed detail, and material unknowns are explicit enough to guide the draft. ## 2. Draft by reader value Apply every relevant rule below. ### Make the point findable - Open with the answer for a question, the outcome for completed work, or the required action for instructions. - Give each paragraph one job and put its point near the beginning. - Use short, descriptive headings when readers may scan or return to the text. - Put prerequisites before the action that needs them and warnings beside the affected step. - Separate the critical path from background, alternatives, and optional improvements. - For long text, provide a concise orientation that identifies the important sections without repeating the full conclusion. ### Write clearly - Use common, concrete words and direct sentence structures. - Keep necessary technical terms. Define an unfamiliar term on first use, then use it consistently. - Expand an unfamiliar acronym on first use. - Make implied relationships explicit: who does what, what causes what, which condition controls an action, and what pronouns refer to. - Use useful repetition when it prevents ambiguity. - Use literal language for instructions, warnings, decisions, and other text where an implied meaning could mislead. Use a metaphor only when it improves understanding and explain it when the reader may not share it. - Use connected prose for explanations, bullets for choices, and numbered lists for steps. - Keep the tone direct, calm, and respectful. Describe the actual effort instead of labeling a task “easy” or “simple.” ### Make action easy - Before a multi-step procedure, state its goal and observable completion result. Include prerequisites, required resources, or a brief overview when the reader needs them. Give an effort or time estimate only when evidence supports it; state the assumptions and use a range when uncertainty matters. - Give each numbered step one bounded action. Include necessary intermediate steps rather than making the reader infer them. - Keep the critical path short. Group branches as conditions: “If X, do Y.” - Show exact commands, values, locations, and expected results where they help the reader act. - When several items compete for attention, separate what to do now from what can wait. - When work remains, end with one obvious next action the reader can start now. When the task is complete, end with the result rather than a ceremonial closing. ### Reduce memory work Keep the information needed for an action on screen and near that action. Prefer explicit nouns over ambiguous references to earlier text. In a continuing exchange, restore only the state needed to resume: what is complete, the current position, and the next move. Make progress visible in concrete terms. Say what now works, what remains blocked, or which artifact changed. Keep tangential discoveries separate from the reader's primary goal. ### Explain in a useful order For a concept, give the direct answer or mental model, then one concrete example, then explain how it works. Put limits, trade-offs, edge cases, and formal detail after the core idea. For instructions, apply the action and memory rules above in execution order. ### Preserve meaning Preserve every needed condition, exception, number, unit, warning, uncertainty, scope boundary, and causal link. Keep a rewrite at the source's level of detail unless the reader asks for a summary. Pair precise wording with a plain explanation when plain wording alone would weaken the claim. When adapting source material, run a separate fidelity pass. Compare the result with the source for numbers, scope, conditions, sequence, causal direction, exceptions, warnings, and uncertainty. For documentation, describe only the supported current workflow and intended state, and use interface and code terms exactly. This step is complete when the draft applies every relevant rule and preserves the required meaning, scope, and detail. ## 3. Verify from the reader's position Before finishing, silently check: - Can the intended reader understand the first paragraph without hidden context? - Is the main point easy to find? - Are unfamiliar terms, pronouns, conditions, and expected results clear? - Is each instruction independently visible, bounded, and complete? - Can the reader identify the next action or the completed result? - Can the reader resume without reconstructing relevant state from earlier messages? - Is every detail useful for the reader's goal? - Can any sentence or section be removed without losing necessary meaning or usability? - Did the text preserve the source's scope and meaning? Keep the checklist internal. Revise until every applicable check passes or report an unresolved constraint explicitly. For a draft of at least 800 words, a fidelity-sensitive adaptation, or a consequential decision, follow [`references/independent-review.md`](references/independent-review.md). For content that claims cognitive accessibility, validate with representative users; model review is comprehension and fidelity evidence, not accessibility evidence. Verification is complete when every applicable check passes, every blocking independent-review finding is resolved, any required representative-user validation is complete, and any remaining constraint is visible to the reader. If representative-user validation is unavailable, remove the cognitive-accessibility claim or label it as unvalidated. ## Foundations When evaluating or changing this skill's principles, consult [`references/foundations.md`](references/foundations.md) for their sources, scope, and authority. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [tschuehly](https://github.com/tschuehly) - **Source:** [tschuehly/skill-incubator](https://github.com/tschuehly/skill-incubator) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-tschuehly-skill-incubator-write-for-humans - Seller: https://agentstack.voostack.com/s/tschuehly - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.