Agent skills for every kind of work.
A skill is a folder with a markdown file in it, holding the order to do something in and the traps worth knowing. 17 of them here, across 11 kinds of work, in the SKILL.md format any agent reads. 12 need nothing installed. Copy one, edit it, keep it.
code quality and testing
Review a diff for what actually matters, ranked by consequence, with confidence marked so a guess is never dressed as a bug.
Read the skillNarrow a bug down instead of guessing at it, and explain the cause before proposing any fix.
Read the skillTests that fail when behaviour breaks, rather than restating the implementation back at itself.
Read the skillChange structure without changing behaviour, in steps that can each be verified and reverted.
Read the skillbackend and apis
Review an HTTP API for the decisions that become permanent once a client depends on them, and say plainly what is breaking.
Read the skilldesign and accessibility
Find the accessibility problems that actually block someone, ranked by impact rather than by rule number.
Read the skillsecurity
A header review that ranks by real risk, rules out obsolete headers, and never proposes a CSP that breaks the site.
Read the skill (needs live checks)Turn a vulnerability scan into an ordered list of actions by working out which findings are actually reachable.
Read the skilldevops and infrastructure
Narrow an outage to a layer before guessing at causes, so nobody changes DNS during an application fault.
Read the skill (needs live checks)network and email
Why mail bounces or lands in spam, read in the order that stops you changing the wrong record.
Read the skill (needs live checks)Certificate faults on a live host, including the browser caching trap that makes a broken chain look healthy.
Read the skill (needs live checks)seo and growth
Whether AI assistants can fetch and read your pages, and the several things people get wrong about the controls.
Read the skill (needs live checks)research and data
Findings traceable to something actually read, with "not found" reported as a result instead of filled in.
Read the skillCheck the data before trusting any number it produces, because most wrong analyses are correct arithmetic on misunderstood rows.
Read the skillwriting and content
A style guide an assistant actually applies, so published text sounds like one person wrote it. Written to be edited.
Read the skillplanning and product
Turn a vague request into something buildable, with assumptions surfaced rather than quietly merged into the requirements.
Read the skillagent architecture
Write a SKILL.md that fires when it should and changes the work when it does. The skill for writing skills.
Read the skillwriting your own
The useful part of a skill is rarely the explanation. It is the order of operations, the traps that are not in a model's general knowledge, and your own conventions. Start with the skill for writing skills, then what agent skills are, how to write your first one and how to test it. If you are not sure whether you need a skill at all, skills compared with MCP servers is the one to read first.
questions
What is an agent skill?
A directory containing a SKILL.md file: YAML frontmatter with a name and a description, then markdown instructions. An assistant reads the description to decide whether the skill is relevant and loads the body only when it is. Any agent that reads SKILL.md can use one.
Where do I put one?
In Claude Code, ~/.claude/skills/<name>/SKILL.md makes it available everywhere, and .claude/skills/<name>/SKILL.md keeps it in a repository so your team gets it on clone. There is no upload step: the files are read from disk.
Do I need to install anything?
For most of these, no. They are instructions only and work with nothing installed. A few look things up on the live internet, which instructions cannot do on their own, and those say so on their page and need an MCP server behind them.
Will a skill make an assistant able to do something new?
No, and this is the most common misunderstanding. A skill changes how a job is approached, not what the assistant can reach. If the gap is access rather than judgement, you need a tool or an MCP server instead. A skill wrapped around a capability that does not exist produces confident invention rather than an error.
Can I change them?
They are meant to be edited, and most of them are only half finished until you do. The order of checks and the traps are the general part. Your own conventions, escalation paths and recurring faults are what make a skill worth keeping, and only you can write those.