# Agent skills Canonical page: https://dominiklukes.net/agents/skills These are the skills Dominik uses in his own work and shares with the people he works with and teaches. They work with Claude Code, Codex and any other agent that reads SKILL.md files. 41 skills, 6 tasks, 5 learning paths. Pinned from [the skills repository](https://github.com/techczech/dominiks-agent-skills) at commit 5ca7a1c. The skills are MIT-licensed. ## Find a skill ### Research with sources (13) Papers, sources, fact-checking and corpora. - [academic-pdf-to-mkd](https://github.com/techczech/dominiks-agent-skills/tree/main/research/academic-pdf-to-mkd): Converts academic PDFs, including scans, into Markdown with figures, tables and page images as separate files. Use it when you want a paper or book chapter in a form an agent can read and quote accurately. - [ai-model-benchmarks](https://github.com/techczech/dominiks-agent-skills/tree/main/research/ai-model-benchmarks): Builds offline comparison charts and tables of AI models from Artificial Analysis, Arena and OpenRouter data. Use it when you need a current chart of model quality, price or usage for a talk or report. Needs: Needs an account. - [anonymise](https://github.com/techczech/dominiks-agent-skills/tree/main/writing/anonymise): Finds and redacts personal information in text on your own machine with an open-weight model, with no cloud service. Use it when you need to remove names and contact details from transcripts or survey responses before sharing them. Needs: Runs on your Mac. - [archive-scout](https://github.com/techczech/dominiks-agent-skills/tree/main/research/archive-scout): Searches, quotes and counts words in your own writing and reading highlights with the local scout tool, and writes cited reports with a JSON evidence file. Use it when you want to find and quote what you have written or highlighted, or count how your use of a word changed over time. - [cesky-narodni-korpus-skill](https://github.com/techczech/cesky-narodni-korpus-skill): Queries the Czech National Corpus (KonText) for concordances, frequencies and collocations from inside an agent. Use it when you study Czech and want corpus evidence without leaving the agent. Needs: Needs an account. - [claim-fact-checker](https://github.com/techczech/dominiks-agent-skills/tree/main/research/claim-fact-checker): Checks every factual claim in a report against its sources, one isolated checker per claim, and writes findings and a fix plan. Use it when you have a report, review or news item and want each claim checked before it goes out. - [corpus-tools-skill](https://github.com/techczech/corpus-tools-skill): Sets up a local corpus-linguistics toolchain on a Mac (NLTK, spaCy, Stanza, Corpus Workbench, OPUS) with a parallel-concordance tool. Draft. Use it when you want to run corpus linguistics on your own machine. Needs: Runs on your Mac. - [paper-reviewer-skill](https://github.com/techczech/dominiks-agent-skills/tree/main/research/paper-reviewer-skill): Imports a paper from Zotero, extracts its text, figures and tables, and writes structured reviews. Use it when you want a structured review of a paper that is in your Zotero library. - [research-translation](https://github.com/techczech/dominiks-agent-skills/tree/main/research/research-translation): Runs auditable translation and back-translation of participant-facing research documents such as consent forms. Use it when you need a consent form or questionnaire translated with a record of each check. - [sketchengine-skill](https://github.com/techczech/sketchengine-skill): Connects an agent to Sketch Engine to run word sketches, word lists and keyword analyses on its corpora. Use it when you have a Sketch Engine subscription and want an agent to run queries. Needs: Needs an account. - [speech-to-text-skill](https://github.com/techczech/speech-to-text-skill): Transcribes audio locally on Apple Silicon (Qwen3-ASR, Parakeet, Whisper) into time-coded transcripts, with optional speaker labels. Use it when you have a recording of a lecture, interview or meeting that needs a transcript. Needs: Runs on your Mac. - [video-analysis](https://github.com/techczech/dominiks-agent-skills/tree/main/media/video-analysis): Describes, summarises or answers questions about a video on your own machine with a local model. Use it when you have a screen recording or video and want a description or timestamped scene list without uploading it. Needs: Runs on your Mac. - [zotero-wrangler](https://github.com/techczech/dominiks-agent-skills/tree/main/research/zotero-wrangler): Manages Zotero imports, linked notes, evidence-backed tags and scoped native full-text discovery. Use it when you want an agent to import references, fetch PDFs or tidy collections in Zotero. ### Write and edit (10) Editing, readability and anonymising. - [archive-scout](https://github.com/techczech/dominiks-agent-skills/tree/main/research/archive-scout): Searches, quotes and counts words in your own writing and reading highlights with the local scout tool, and writes cited reports with a JSON evidence file. Use it when you want to find and quote what you have written or highlighted, or count how your use of a word changed over time. - [claim-fact-checker](https://github.com/techczech/dominiks-agent-skills/tree/main/research/claim-fact-checker): Checks every factual claim in a report against its sources, one isolated checker per claim, and writes findings and a fix plan. Use it when you have a report, review or news item and want each claim checked before it goes out. - [doc-editorial-skill](https://github.com/techczech/dominiks-agent-skills/tree/main/writing/doc-editorial-skill): Audits, reviews and fact-checks markdown documentation, proposing each change and asking before editing. Use it when you have a draft or documentation set and want structured editing help. - [expository-writing](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/expository-writing): Keeps an agent's prose literal and specific: every sentence says something checkable. Use it when you want an agent's reports, documentation or web copy to be plain and specific. - [metaphor](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/metaphor): Helps an agent use and examine metaphors without over-extending them. Use it when you are explaining something with a metaphor, or using one to frame a design. - [readability-skill](https://github.com/techczech/readability-skill): Helps write more readable text and design more readable slides. Use it when you want text or slides that are easier to read, including for dyslexic readers. - [research-translation](https://github.com/techczech/dominiks-agent-skills/tree/main/research/research-translation): Runs auditable translation and back-translation of participant-facing research documents such as consent forms. Use it when you need a consent form or questionnaire translated with a record of each check. - [single-html-document](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/single-html-document): Builds self-contained single-file HTML documents (reports, explainers, slide decks, feedback forms) that work offline. Use it when you want one HTML file you can email, open from disk or put on a web server. - [writeflex](https://github.com/techczech/dominiks-agent-skills/tree/main/writing/writeflex): Lets agents read, organise and prepare Markdown content for the WriteFlex app. Use it when you keep your writing in WriteFlex and want an agent to help with collections and documents. Needs: Needs one of Dominik's apps. - [writing-modes](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/writing-modes): Makes an agent decide who a text is for and in whose name it speaks before it writes, and load only that register's rules. Use it when you want an agent to settle who a text is for before it writes, so that chat replies, instruction files and anything sent in your name each get the right treatment. ### Teach and present (7) Presentations and learning materials. - [ai-model-benchmarks](https://github.com/techczech/dominiks-agent-skills/tree/main/research/ai-model-benchmarks): Builds offline comparison charts and tables of AI models from Artificial Analysis, Arena and OpenRouter data. Use it when you need a current chart of model quality, price or usage for a talk or report. Needs: Needs an account. - [metaphor](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/metaphor): Helps an agent use and examine metaphors without over-extending them. Use it when you are explaining something with a metaphor, or using one to frame a design. - [readability-skill](https://github.com/techczech/readability-skill): Helps write more readable text and design more readable slides. Use it when you want text or slides that are easier to read, including for dyslexic readers. - [single-html-document](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/single-html-document): Builds self-contained single-file HTML documents (reports, explainers, slide decks, feedback forms) that work offline. Use it when you want one HTML file you can email, open from disk or put on a web server. - [talkweaver](https://github.com/techczech/dominiks-agent-skills/tree/main/teaching/talkweaver): Lets agents write and revise presentation content for the TalkWeaver app, including converting PowerPoint files and processing recordings. Use it when you build talks in TalkWeaver and want help with outlines, handouts or recordings. Needs: Needs one of Dominik's apps. - [text-as-graph](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/text-as-graph): Treats documents and courses as connected parts so they can be reused and recombined without copies drifting. Use it when you are designing how documents, courses or slides are structured and reused. - [text-to-image-skill](https://github.com/techczech/text-to-image-skill): Generates and edits images locally on Apple Silicon with open models such as FLUX.2 Klein and Qwen-Image. Use it when you want illustrations for slides or documents without sending prompts to a service. Needs: Runs on your Mac. ### Build software and documents (14) Software, design and HTML documents. For planning before you build, these pair well with [Matt Pocock's skills](https://github.com/mattpocock/skills): planning skills for grilling an idea, writing a spec and splitting the work into tickets. - [agent-roles](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/agent-roles): Orchestrator playbook for Claude Code: which model plans, builds, fixes, reviews and tests, with dated seat choices, cost evidence and paste-in briefs. Use it when you run Claude Code as the planner and hand building, fixing, review and testing to other models. - [app-design-elements](https://github.com/techczech/app-design-elements): The shared elements every desktop app carries: commands, keyboard, menus, Settings and agent access. Use it when you want an agent to build a desktop app whose commands, shortcuts, menus and Settings window work the same way as in your other apps. - [codex-images-workflow](https://github.com/techczech/dominiks-agent-skills/tree/main/media/codex-images-workflow): Treats generated images as reproducible files: a written brief first, the image beside it, both versioned. Use it when you generate posters, mockups or diagrams and need to regenerate or revise them later. - [cognitive-walkthrough](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/cognitive-walkthrough): Runs a cognitive walkthrough on a mockup, screenshot or live page to check how easily a new user can learn it, and writes a learnability report. Use it when you have a design or a prototype and want to know where a first-time user would get stuck. - [design-first-dev](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/design-first-dev): Takes an app or feature from user journeys and drawn designs to a locked mockup before any code is written. Use it when you are about to build an app or a feature and want the design settled before any code is written. - [dev-traffic-control](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/dev-traffic-control): Lets agents file test requests and reviews in Dev Traffic Control, a desktop review app, and collect the answers. Use it when you review agent work in Dev Traffic Control and want agents to file requests and read your answers. Needs: Needs one of Dominik's apps. - [goalify](https://github.com/techczech/goalify): Turns a project or product interview into durable goal prompts for OpenAI Codex. Use it when you want to hand Codex a long-running goal that is written down clearly. - [image-to-design](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/image-to-design): Takes a design from a written brief through generated mockups to HTML/CSS, keeping every prompt, image and revision on disk. Use it when you are exploring a user interface visually and want each version kept and comparable. - [judgement-vs-code](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/judgement-vs-code): Helps an agent decide when a task needs reading and judgement rather than string matching. Use it when your agent is about to write code that processes language or document content. - [project-changelog](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/project-changelog): Keeps a structured changelog folder in a repository alongside its git history. Use it when you want a project's history, decisions and backlog recorded in the repository itself. - [single-html-document](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/single-html-document): Builds self-contained single-file HTML documents (reports, explainers, slide decks, feedback forms) that work offline. Use it when you want one HTML file you can email, open from disk or put on a web server. - [skill-explorer](https://github.com/techczech/skill-explorer): Turns a SKILL.md file into an interactive web page showing its structure, code and connections. Use it when you want to understand a skill someone else wrote, or explain one to a class. - [text-as-graph](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/text-as-graph): Treats documents and courses as connected parts so they can be reused and recombined without copies drifting. Use it when you are designing how documents, courses or slides are structured and reused. - [ux-journeys](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/ux-journeys): Pins down who the user is, what they are trying to do and what they see at each step, before visual design starts. Use it when you are starting design work and need to agree who the user is and what they do at each step. ### Work with images, audio and video (6) Generate, transcribe and describe media. - [codex-images-workflow](https://github.com/techczech/dominiks-agent-skills/tree/main/media/codex-images-workflow): Treats generated images as reproducible files: a written brief first, the image beside it, both versioned. Use it when you generate posters, mockups or diagrams and need to regenerate or revise them later. - [image-to-design](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/image-to-design): Takes a design from a written brief through generated mockups to HTML/CSS, keeping every prompt, image and revision on disk. Use it when you are exploring a user interface visually and want each version kept and comparable. - [speech-to-text-skill](https://github.com/techczech/speech-to-text-skill): Transcribes audio locally on Apple Silicon (Qwen3-ASR, Parakeet, Whisper) into time-coded transcripts, with optional speaker labels. Use it when you have a recording of a lecture, interview or meeting that needs a transcript. Needs: Runs on your Mac. - [text-to-image-skill](https://github.com/techczech/text-to-image-skill): Generates and edits images locally on Apple Silicon with open models such as FLUX.2 Klein and Qwen-Image. Use it when you want illustrations for slides or documents without sending prompts to a service. Needs: Runs on your Mac. - [text-to-speech-skill](https://github.com/techczech/text-to-speech-skill): Generates speech locally on Apple Silicon with several open voice models, including voice cloning and Czech. Use it when you want narration or an audio version of a text. Needs: Runs on your Mac. - [video-analysis](https://github.com/techczech/dominiks-agent-skills/tree/main/media/video-analysis): Describes, summarises or answers questions about a video on your own machine with a local model. Use it when you have a screen recording or video and want a description or timestamped scene list without uploading it. Needs: Runs on your Mac. ### Set up your agents (9) Instructions and how you work with AI. - [agent-roles](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/agent-roles): Orchestrator playbook for Claude Code: which model plans, builds, fixes, reviews and tests, with dated seat choices, cost evidence and paste-in briefs. Use it when you run Claude Code as the planner and hand building, fixing, review and testing to other models. - [agents-md-streamline](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/agents-md-streamline): Writes and tightens AGENTS.md and CLAUDE.md files so they tell an agent what to do, not the project's history. Use it when your instruction files have grown long and the agent follows them less well. - [ai-working-style-builder](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/ai-working-style-builder): Helps you describe how you work so AI tools fit your communication, attention and record-keeping habits. Use it when you are starting to work with an agent and want it to fit how you think and write. [Interactive guide](https://techczech.github.io/dominiks-agent-skills/ai-interaction-builder/). - [coordination-hub](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/coordination-hub): Sets up a hub repo where every agent session starts, holding a registry of your projects, a backlog, handoff notes and a lessons log, with a routine check for uncommitted or unpushed work. Use it when you work on many projects with agents and cannot keep track of where each one lives or what state you left it in. - [expository-writing](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/expository-writing): Keeps an agent's prose literal and specific: every sentence says something checkable. Use it when you want an agent's reports, documentation or web copy to be plain and specific. - [global-agent-instructions](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/global-agent-instructions): Keeps one global instruction file and installs it into Codex, Claude Code, Gemini CLI, OpenCode and pi. Use it when you use more than one agent and want the same instructions in each. - [judgement-vs-code](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/judgement-vs-code): Helps an agent decide when a task needs reading and judgement rather than string matching. Use it when your agent is about to write code that processes language or document content. - [learn-from-dominiks-setup](https://github.com/techczech/dominiks-agent-skills/tree/main/estate/learn-from-dominiks-setup): Looks at your own agent setup, compares it with Dominik's practices and points you to the few skills worth trying; you start by pasting one line into your agent, with nothing to install first. Use it when you want to know which of Dominik's skills are worth trying and would rather your agent looked at your setup than read a long list. - [writing-modes](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/writing-modes): Makes an agent decide who a text is for and in whose name it speaks before it writes, and load only that register's rules. Use it when you want an agent to settle who a text is for before it writes, so that chat replies, instruction files and anything sent in your name each get the right treatment. ### What each skill needs - Runs on your Mac (6): anonymise, corpus-tools-skill, speech-to-text-skill, text-to-image-skill, text-to-speech-skill, video-analysis - Needs one of Dominik's apps (3): dev-traffic-control, talkweaver, writeflex - Needs an account (3): ai-model-benchmarks, cesky-narodni-korpus-skill, sketchengine-skill ## What is a skill? - **What a skill is.** A skill is a folder with instructions for agents on how to perform certain tasks consistently. - **What is in the folder.** The folder contains a `SKILL.md` file that describes what the skill does and links to other instructions in a `references/` folder or to scripts the agent can run in a `scripts/` folder. Some skills have only a single file; others are extensive collections of instructions and scripts, and some also contain `assets/` such as images. Some skills depend on other apps being installed on your machine. - **Which agents and models.** Agent skills are a universal standard and work with any desktop agent, and now with some chatbots. Some skills have instructions that assume the capabilities of a certain model or agent. They will still work with any model, but the model may be less reliable on those tasks. For example, the image-to-design skill assumes the agent can generate images; at the time of writing, Codex and the ChatGPT desktop app can do this. - **Review your skills over time.** As the models behind agents get better, the need for some skills decreases because the models can already make good decisions. A skill can even become counterproductive when it goes against the agent's ability to make good decisions. Review your skills periodically to make sure they still serve their intended purpose. ## How to install a skill To install a skill means registering it with your agent, so it can use the skill when needed, and putting it in a folder on your computer that the agent can access. You can install a skill as it is, ask the agent to modify it for your purposes, or have it build a new skill to fit your workflow. There are three ways to install one. ### 1. Ask your agent to choose Open Claude Code, Codex or another agent and paste this prompt. Replace the … with a sentence about your work. The agent reads the repository, suggests skills and installs the ones you agree to. ```text Look at https://github.com/techczech/dominiks-agent-skills and tell me which of these skills would fit my work. I mainly … Install the ones I pick. ``` ### 2. Install with the skills command Run this in a terminal. It lists the skills in the repository and lets you choose which ones to install. You need Node.js. ```sh npx skills add techczech/dominiks-agent-skills ``` ### 3. Copy the skill’s folder Download the repository and copy one skill’s folder into your agent’s skills folder. For Claude Code that is ~/.claude/skills/. Skills marked ‘own repository’ are copied from their own repository in the same way. ```sh git clone https://github.com/techczech/dominiks-agent-skills cp -r dominiks-agent-skills// ~/.claude/skills/ ``` ### Check before you install Always ask your agent to review a skill before you use it, to make sure it has no hidden instructions and no scripts that have bugs or are malicious. ## Learning paths Each path is a short sequence in the order Dominik teaches it. Try the steps one at a time; each one gives you something to use in the next. ### Start here Set up your agent, then make your first document. 1. [learn-from-dominiks-setup](https://dominiklukes.net/agents/skills/learn-from-dominiks-setup): Paste one line into your agent and it compares your setup with Dominik's and suggests where to start. 2. [ai-working-style-builder](https://dominiklukes.net/agents/skills/ai-working-style-builder): Describe how you work and what you need from an AI tool. 3. [agents-md-streamline](https://dominiklukes.net/agents/skills/agents-md-streamline): Turn that into a short instruction file your agent will follow. 4. [global-agent-instructions](https://dominiklukes.net/agents/skills/global-agent-instructions): Install the same instructions in every agent you use. 5. [coordination-hub](https://dominiklukes.net/agents/skills/coordination-hub): Set up one hub where every agent session starts and finds your projects. 6. [single-html-document](https://dominiklukes.net/agents/skills/single-html-document): Then make something you can see: ask for a one-page explainer and open it in your browser. ### For researchers From a PDF to a checked review. 1. [academic-pdf-to-mkd](https://dominiklukes.net/agents/skills/academic-pdf-to-mkd): Turn the PDF into Markdown so the agent can read its text, tables and figures. 2. [zotero-wrangler](https://dominiklukes.net/agents/skills/zotero-wrangler): File the paper and your notes in your Zotero library. 3. [paper-reviewer-skill](https://dominiklukes.net/agents/skills/paper-reviewer-skill): Write structured reviews of a paper from your library. 4. [claim-fact-checker](https://dominiklukes.net/agents/skills/claim-fact-checker): Check the claims in your own write-up against its sources. 5. [anonymise](https://dominiklukes.net/agents/skills/anonymise): Remove names and contact details from transcripts or survey responses before you share them. 6. [ai-model-benchmarks](https://dominiklukes.net/agents/skills/ai-model-benchmarks): Only if your work compares AI models: build the charts from current data. ### For teachers From a topic to slides and handouts people can read. 1. [readability-skill](https://dominiklukes.net/agents/skills/readability-skill): Write the text so that everyone in the room can read it. 2. [single-html-document](https://dominiklukes.net/agents/skills/single-html-document): Package a handout or explainer as one file that students can open offline. 3. [talkweaver](https://dominiklukes.net/agents/skills/talkweaver): Move the talk itself into the TalkWeaver app. 4. [text-to-image-skill](https://dominiklukes.net/agents/skills/text-to-image-skill): Add images last, when you know what each slide needs. ### For writers Draft, edit, check and share. 1. [doc-editorial-skill](https://dominiklukes.net/agents/skills/doc-editorial-skill): Fix the structure first, before you polish sentences you might cut. 2. [readability-skill](https://dominiklukes.net/agents/skills/readability-skill): Then make the sentences and the layout easier to read. 3. [claim-fact-checker](https://dominiklukes.net/agents/skills/claim-fact-checker): Check every factual claim against its sources. 4. [writeflex](https://dominiklukes.net/agents/skills/writeflex): Keep your drafts and collections in the WriteFlex app, where an agent can read and organise them with you. ### For people building apps From an idea to a checked prototype. 1. [ux-journeys](https://dominiklukes.net/agents/skills/ux-journeys): Write down who the user is and what they do at each step. 2. [design-first-dev](https://dominiklukes.net/agents/skills/design-first-dev): Take those journeys through drawn designs to a locked mockup before any code. 3. [image-to-design](https://dominiklukes.net/agents/skills/image-to-design): Explore what the app could look like with generated mockups. 4. [cognitive-walkthrough](https://dominiklukes.net/agents/skills/cognitive-walkthrough): Walk through the design as a first-time user to find where people get stuck. 5. [single-html-document](https://dominiklukes.net/agents/skills/single-html-document): Write up the plan or a report as one file anyone can open. 6. [project-changelog](https://dominiklukes.net/agents/skills/project-changelog): Record what changed and why as you build. 7. [dev-traffic-control](https://dominiklukes.net/agents/skills/dev-traffic-control): Ask people to test the result and collect their answers in one place. Before step 1, plan the app with [Matt Pocock's skills](https://github.com/mattpocock/skills). ## Philosophy for agents On 17 July 2026 Dominik started dictating conversations about how he thinks about language and meaning. He has a background in cognitive linguistics and metaphor research, and he wanted agents to work from the ideas he uses himself rather than from their defaults. Each topic became an instruction file that agents load, with an essay version for people. A model drafted each distillation, a model from a different family reviewed it independently, and Dominik decided every disputed point. The essays were drafted alongside and are not yet published. - **Grep and Grok** ([judgement-vs-code](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/judgement-vs-code)): Says when a task over language needs reading and judgement (Grok) and when string operations (Grep) are right; the failure is using one where the other belongs. The core line from the series is "we think in schemas and we speak in propositions", and the name comes from Dominik's Grep and Grok cartoons. - **Text as a graph** ([text-as-graph](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/text-as-graph)): Treats documents and courses as graphs of connected parts. Slides, handouts and outlines are views derived from them, and every copy records where it came from and how to regenerate it. This was the first file in the series and set the pattern for the rest. - **Metaphor** ([metaphor](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/metaphor)): Metaphor mappings are partial and negotiated, so an agent should neither dismiss a metaphor nor import everything from its source. It draws on Dominik's metaphor research (Metaphor Hacker blog). - **Expository writing** ([expository-writing](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/expository-writing)): Every sentence should say something a reader can use or dispute, with no decorative or journalistic language unless asked. On 24 July 2026 a live correction added that register (how sentences are written) and structure (layout) are separate axes, and a complaint about one is not licence to change the other. - **Writing modes** ([writing-modes](https://github.com/techczech/dominiks-agent-skills/tree/main/philosophy-for-agents/writing-modes)): Before writing any prose, the agent decides who the text is for and in whose name it speaks: for agents (instruction files, logs), for you (chat, READMEs, explanations), or as you (anything sent or published under your name, using only your own sources and voice). It announces the register and reads only that register's rules. It came out of the same series as Expository writing: Dominik renamed the registers on 24 July 2026 after correcting an agent that had announced a register without applying it. You can rename the registers, add one, or point the 'as you' register at your own voice notes. ## Works with Dominik's apps These skills let an agent work inside apps Dominik makes. Each one needs its app installed. - [Dev Traffic Control](https://techczech.github.io/dev-traffic-control/): A Mac app where your coding agents send you checks, design reviews and handoffs, and you answer them. Skill: [dev-traffic-control](https://github.com/techczech/dominiks-agent-skills/tree/main/dev/dev-traffic-control). - [WriteFlex](https://writeflex.app): A Markdown writing app for collections of documents. Skill: [writeflex](https://github.com/techczech/dominiks-agent-skills/tree/main/writing/writeflex). - [TalkWeaver](https://github.com/techczech/talkweaver-app): An app for writing, presenting and recording talks. Skill: [talkweaver](https://github.com/techczech/dominiks-agent-skills/tree/main/teaching/talkweaver).