agent-skills

Overview

agent-skills is a library of ready instructions for AI coding agents. Instead of explaining your process in every session, you install it once and call a skill by name.

It stays independent: Skribly is only one of its integrations, and it works in any project.

Workflow

Run a task from a ticket to a commit, write commits, save work state and split large jobs.

Quality

Reviews, backend and end-to-end tests, a capped test-fixing loop and debugging.

Audits

PageSpeed, accessibility, SEO and security checks with concrete fixes.

Operations

Database migrations, Docker, CI, dependency upgrades and release notes.

Token saving

Skills load on demand, and scripts condense logs and reports before a model reads them.

Your OK first

Commits and comments wait for your approval, and expensive checks are proposed, not run silently.

How you use it

• agent-skills install --target all
• /task 123
• /commit
• /wcag-audit

Categories

Workflow (7)

Move a task from a ticket to a commit, and keep the project around it healthy.

Quality and tests (11)

Find problems before they reach a commit.

Performance (2)

Measure first, then prove the gain.

Security and accessibility (3)

Check for what should never ship.

Docs (3)

Keep what people and agents read up to date.

Meta (1)

Rules for how an agent works.

Browse all 27 skills

Workflows

From a ticket to a commit

/task 482 fetches the issue by script, analyses it, creates a branch after your approval and works in stages. It commits after your OK and never pushes.

More ways to call a task

A task can start from a number, a key, a URL or a sentence, and it can also estimate, plan or draft a comment.

/task 482
/task SHOP-42
/task fix the date sorting
/task 482 estimate
/task 482 plan
/task 482 pr
/task 482 comment

Continue in another agent

$handoff resume 482 continues in Codex what Claude started, because the state lives in a file and not in a transcript.

Work with a team

Install at project scope and commit .claude/skills and .agents/skills, so everyone gets the same skills. .tasks/ stays personal.

Read the agent-skills guide

Safety

• Commits wait for your OK, and skills never push, merge, rebase or force unless you ask.
• The tracker command is read-only unless you publish a comment with --post --yes.
• Credentials stay outside the repository, and tokens go only to the tracker's own host.
• Files you did not install, or edited by hand, are never overwritten without --force.
• Your code is not collected or uploaded.

See how skills save tokens

Details

Screenshots

Open larger image: Skill catalogue