personality
Draft, edit, or review professional writing in the user's conversational engineering voice.
Skill by ahstn
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/ahstn-personality
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: writing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/personality
What's inside
- Skill name
personality- Description
- Draft, edit, or review professional writing in the user's conversational engineering voice. Supports close preservation, clearer revisions, audience adaptation, personal reflections, and evidence-based style calibration.
- License
- MIT
- Instructions
- 77 lines in SKILL.md · view on GitHub
Agent Skills spec issues
- files: referenced but missing: references/personal-voice.md, references/guidance-direct-messages.md, references/guidance-blog-posts.md, references/guidance-professional-formats.md, references/guidance-reflections.md
Files (1)
- SKILL.md
SKILL.md (first lines)
# Personality and voice Preserve the writer's judgment, warmth, and useful edge while making the writing easier to understand. Match the publication and the requested degree of editing. A personal voice includes how the writer reasons and relates to readers, as well as vocabulary and rhythm. ## Precedence 1. Preserve meaning, facts, evidence, uncertainty, ownership, commitments, technical identifiers, and any limits on what may be included. 2. Follow the user's explicit requirements for audience, purpose, publication, tone, format, length, and editing mode. 3. Apply the selected publication's requirements. 4. Apply the personal profile and the selected improvement goals. 5. Use shared editing defaults and quantitative diagnostics. The personal profile overrides generic voice defaults only. Examples do not override the task or authorise new facts, experiences, opinions, deadlines, or promises. Flag material inconsistencies instead of silently repairing them through invention. ## Workflow and modes Identify the operation: draft, edit, review, or analyse. Infer the audience and purpose when clear, including whether the text informs, explores, requests, or records a decision. Ask a focused question only when missing information would materially change the result; continue independent work when possible. When working from mixed notes, distinguish the writer's text from quotations, feedback, prompts, suggested wording, and excluded passages. Follow the user's current inclusion instructions; do not silently promote background notes or someone else's judgment into the writer's claims. Read [personal voice](references/personal-voice.md), then the relevant publication reference below. Choose the editing mode from the user's request: | Mode | Use when | Editing behaviour | |---|---|---| | Preserve | "Light edit", "keep my wording", or a close copy edit | Repair errors and unclear passages. Keep clear sentence shapes, structure, personal asides, and degree of formality. | | Refine | Default for drafting or editing without another mode | Keep the writer's stance and warmth. Untangle sentences, tighten evidence, remove repetition, and improve organisation where it helps the reader. | | Adapt | The user asks for another audience, register, or desired style | Change presentation to serve that purpose. Retain the factual and personal boundaries above. Borrow named qualities, such as clarity or narrative pacing, without inventing a new identity. | Publication and mode are separate choices: a performance review can receive a light edit; a short chat can need substantial clarification. When reviewing without a rewrite, return findings and targeted suggestions. Do not produce a rewritten artifact unless requested. For a rewrite, identify a few voice signals to keep before editing. Keep this working note internal. Return the requested artifact; explain changes only when requested or when a substantial reorganisation needs explanation. Do not append a standard "What changed" section to every response. ## Publication router Read one primary reference. Read another only when the artifact has a second purpose that needs it. | Publication or purpose | Reference | |---|---| | Chat, Slack, Teams, quick operational update | [Direct messages](references/guidance-direct-messages.md) | | Blog post, engineering article, public essay | [Blog posts](references/guidance-blog-posts.md) | | PR review, email, memo, technical documentation, ADR or RFC | [Professional formats](references/guidance-professional-formats.md), using only the relevant section | | Self-review, performance reflection, growth plan, reflective notes | [Reflections and performance reviews](references/guidance-reflections.md) | Read [annotated examples](references/annotated-examples.md) when calibrating a mode or resolving a likely loss of voice. Read [storytelling and metaphor](references/guidance-storytelling-metaphor.md) when requested or when an analogy would materially clarify a difficult concept. Read [evaluation](references/evaluation.md) for profile extraction, comparison, measurement, or changes to this skill. Required references travel with this skill. If one is unavailable, use the core rules and available guidance, and disclose the limitation when it prevents the requested task. Do not claim a personal match from a profile you could not read. Local source records and the original Obsidian files are optional evidence for calibration, not runtime dependencies. ## Shared editing defaults - Use common words around exact domain terms. Keep names and terms stable; do not rotate synonyms for variety. - Follow the stated goals of clarity, simplicity, brevity, and humanity. Preserve a useful personal aside, admission, or story when it serves the piece. - Keep direct judgments where the writer has made them. Preserve uncertainty that marks a real limit. Remove empty hedging without changing the strength of a claim. - Keep specific warmth, pride, gratitude, disappointment, and earned enthusiasm when relevant to the purpose. These can carry meaning without a metric. Preserve feelings as the writer's experience, not proof of another person's motives. Remove promotional claims that lack support. - Use "I" for the writer's own actions, judgments, and limits; use "we" for shared work. Do not invent consensus or turn help with a task into sole ownership. - Keep responsibility aligned with control. Preserve what the writer can do, what needs another owner or decision, and any capacity limit. Do not turn a shared constraint into a personal failing or an offer to absorb more work. - Keep bluntness or humour when appropriate to the task and source. Do not add contempt, personal attacks, or invented motives. A firm decision does not need to become a tentative suggestion. - Use active voice when the actor matters. Passive voice can help describe system behaviour without accusing a person. - Vary sentence length with the reasoning. Keep a clear longer sentence when its conditions belong together. Split overloaded sentences without turning connected prose into stacked fragments. - Select context and evidence for what this audience needs to understand or assess. Make missing links between ideas explicit. Keep a short sequence of situation, action, and consequence when it explains the point; do not impose that structure on every message. - Cut empty preambles, repeated conclusions, faux-insight setups, ornamental binary contrasts, and decorative lists of three. Preserve a real question or tension when both sides affect the meaning, such as progress alongside an unresolved cost. - Let an unresolved concern remain unresolved when that serves the purpose. Do not add a positive lesson, neat resolution, or new commitment to make an ending feel complete. Include an action or decision when the request calls for one, using only supported details. - Treat words such as "delve", "foster", "empower", "game changer", and "transformative" as prompts to inspect the claim, not proof of authorship. Remove inflated or formulaic uses. Preserve literal technical uses, such as a canary deployment or an agent harness, and quoted wording or identifiers. - Do not add mistakes, profanity, slang, or catchphrases to simulate authenticity. ## Target mechanics - Use plain technical English and UK spelling in authored prose; preserve official names, quotes, code, and identifiers. Follow the user's ASD-STE100 and Flesch-Kincaid grade 7-10 requirements when applicable. A grade score alone does not establish STE compliance. - For sufficiently long prose, use readability results to locate difficult passages. Rewrite around necessary technical terms without changing them. Do not pad easy sentences to reach a grade floor or claim a measured score without running a measure. - Use straight quotation marks and avoid em dashes in authored prose. These are target preferences even where older samples differ. Preserve exact source quotations when fidelity is required. - Use headings, lists, tables, and code because the content needs them. Keep emphasis sparse. Familiar chat may retain an appropriate emoji from the source or established context; do not add decorative emoji by default. - Do not hard wrap Markdown paragraphs at 80 columns. ## Final check Check meaning and commitments first, then publication fit, voice, clarity, and presentation. Ensure that the result answers the actual request, preserves material limits and credit, and contains no invented evidence or generic summary. Remove edits that only make the prose more uniform.
From ahstn/dotfiles, MIT.
More from ahstn/dotfiles
- 0.609score
kubes-mode-v2
ahstn
Personal working mode for software engineering tasks.
SkillAFreshness: A★ 3Claude Code +9 - 0.609score
jira-manager
Jira skill maintainers
Use this skill for all Jira workflows: create epics and issues, transition tickets, add comments, search with JQL, and manage the backlog.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
code-review-and-quality
ahstn
Reviews staged or unstaged diffs, branches, and GitHub pull requests for defects and material code-health costs.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
computer-use-cua-driver
ahstn
Operate native macOS, Windows, or Linux GUI applications with Cua Driver through an available MCP connection or the installed CLI.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
tty-fork
ahstn
Start a fresh or forked OMP session in a right or below split, or a new tab, in tty7, Herdr, or Tern.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
github-librarian
ahstn
Use when looking up GitHub repositories, upstream source, or library internals, either via quick `gh search` queries or by cloning into ~/.cache/github/ for…
SkillAFreshness: A★ 3Claude Code +9 - 0.603score
slack-message
ahstn
Rules for sending Slack messages and reference request payloads.
SkillAFreshness: A★ 3Claude Code +9 - 0.560score
kubes-mode
ahstn
No description from the publisher yet.
SkillAFreshness: A★ 3Claude Code +9 - 0.534score
github-yeet
OpenAI
Use only when the user explicitly asks to stage, commit, push, and open a GitHub pull request
SkillAFreshness: A★ 3Claude Code +9 - 0.514score
code-structure-and-modularity
ahstn
Code structure and module-boundary rubric for writing, extending, reviewing, or reorganizing code.
SkillCFreshness: C★ 3Claude Code +9 - 0.486score
wayfinder
ahstn
Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets on your issue tracker, and resolve them one at a…
SkillCFreshness: C★ 3Claude Code +9 - 0.484score
grill
ahstn
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree.
SkillCFreshness: C★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 12, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/ahstn/dotfiles
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from ahstn
- 0.609score
kubes-mode-v2
ahstn
Personal working mode for software engineering tasks.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
code-review-and-quality
ahstn
Reviews staged or unstaged diffs, branches, and GitHub pull requests for defects and material code-health costs.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
computer-use-cua-driver
ahstn
Operate native macOS, Windows, or Linux GUI applications with Cua Driver through an available MCP connection or the installed CLI.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
github-librarian
ahstn
Use when looking up GitHub repositories, upstream source, or library internals, either via quick `gh search` queries or by cloning into ~/.cache/github/ for…
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
tty-fork
ahstn
Start a fresh or forked OMP session in a right or below split, or a new tab, in tty7, Herdr, or Tern.
SkillAFreshness: A★ 3Claude Code +9 - 0.603score
slack-message
ahstn
Rules for sending Slack messages and reference request payloads.
SkillAFreshness: A★ 3Claude Code +9
Cite this page
personality skill. Agent Mart. https://agentmart.dev/skills/ahstn-personality (last updated 2026-10-24)