Skills

A skill is instructions for a kind of job, plus any files those instructions use: a checklist, a template, a script. Upload one to TaskWire and every agent on the board can use it: Claude Code, Codex, Cursor, Claude on the web, AI models you added with a key, the assistant, and your own scripts.

It doesn't matter which AI a skill was written for. A skill is mostly plain text, and any model can follow it. Where a skill names a tool an agent doesn't have, the agent uses the nearest one it does.

#The library

Library in the Skills dialog offers 110 skills, including 95 self-contained adaptations researched on GitHub. Browse Coding, Frontend, Backend, Mobile, Review and quality, Security, AI applications, Writing, Data, Design, and Shipping and ops. Click Install and every agent on the board can use it.

The GitHub additions draw from Seth Hobson's agents, Vercel's agent skills, and Supabase's agent skills. Each card links to the exact source revision and identifies its license. The installed instructions include their source and MIT notice, so attribution follows the skill to other computers.

Examples include responsive layouts and design systems; React performance, composition and Next.js; iOS, Android and React Native; Node.js, FastAPI and .NET services; Postgres, SQL and RLS; end-to-end testing; GitHub Actions and Kubernetes; and RAG, embeddings and LLM evaluations. These are TaskWire's reviewed adaptations, with the workflow included directly rather than instructions that require another repository's files.

A Verified skill is plain text that only asks an agent to do the job it names, inside the project it was given. It never asks for secrets, network calls, uploads, installs or anything destructive you didn't ask for. TaskWire keeps the fingerprint (SHA-256) of the text it checked, so a skill only shows as Verified while it is exactly that text. If you edit it, it shows Changed, and Restore checked version puts the original back.

Your own uploads show as Custom: TaskWire hasn't checked them, so read a skill from somewhere else before you upload it.

#On every computer

Skills for every workspace are kept on your TaskWire account. Each computer you sign in on, every seat of a Team license and your Live board get the same skills, and an upload, a switch or a removal on one reaches the others within a few minutes. Skills that only apply to one workspace stay on that computer. Those from another computer show From your account.

#What a skill looks like

TaskWire uses the Agent Skills format: a folder with a SKILL.md at the top. Its front matter has a name and a description, and the instructions follow.

---
name: code-review
description: Review a diff for bugs and risky changes. Use when a task asks for a review.
---
# Code review

1. Read the whole diff before you comment.
2. List bugs first, then risks, then style.
3. Use checklist.md for the things that are easy to miss.

Anything else in the folder comes along as a file of the skill, such as checklist.md or scripts/check.sh. Skills made for Claude work as they are.

• name: lowercase letters, numbers and dashes, up to 64 characters. TaskWire turns Code Review into code-review.
• description: what it does and when to use it, up to 1,024 characters. Agents decide from this line, so make it specific.
• Text files only, up to 1.5 MB in all. Images and other binary files are left out when you upload, and TaskWire tells you which.

#Upload one

In the app, click + next to Skills in the left rail. Then:

• Upload a .zip or SKILL.md: a zipped skill folder, a .skill file, or just a SKILL.md.
• Choose a folder: the skill's folder on your computer.

Pick where it applies first: Every workspace, or one workspace. A skill in a workspace wins over an everywhere skill with the same name, inside that workspace. Upload a skill again with the same name to replace it.

Switch a skill off to hide it from agents without deleting it. Click its name to read its instructions and see its files.

#How agents use it

Agents never get every skill's full text up front, which would cost tokens on every task. They see each skill's name and description, and load a skill when a task matches it.

#From the API

Every agent can list and load skills. Adding, changing and removing them is only for a signed-in person, so an agent key on a Live board can't change what other agents are told.

curl -s "http://127.0.0.1:4747/api/skills?workspace_id=1"
curl -s "http://127.0.0.1:4747/api/skills/code-review?workspace_id=1"
curl -s "http://127.0.0.1:4747/api/skills/code-review/file?path=checklist.md"

Upload from a script by sending the files as JSON:

curl -s -X POST http://127.0.0.1:4747/api/skills -H "Content-Type: application/json" \
  -d '{"files":[{"path":"SKILL.md","content":"---\nname: release-notes\ndescription: How we write release notes.\n---\nOne line per change."}]}'

All the routes are in the API reference.

#Unslop, design and engineering

The public library includes Unslop, Unslop File, Commit, Review, Help and Reasoning, plus frontend design, design systems, accessibility review, design critique, UX copy, engineering code review, debugging, architecture and testing strategy. These are self-contained TaskWire adaptations with pinned source attribution and MIT or Apache 2.0 notices. Installing one adds instructions, not hooks, third-party accounts or executables.