When your failing test is still open at lunch, choose a skill that writes evidence rather than discussing it. A skill is a folder of instructions the model opens when it decides it needs them. The catalogue records 13,825 testing skills. Choose one that leaves you with something to check.
When a testing skill reaches the session
The context is everything the model can see while it answers you. A writing skill should turn a requirement into test code, fixtures, assertions, or a repeatable check. A running skill should guide the commands and checks that show whether the software works.
The catalogue has 13,476 skills that load every session. Another 349 wait for a match. That distinction matters when you want help for one testing task.
What test skills cost before you ask
A token is the unit your AI plan bills by, about three quarters of a word. Among skills that load every session, the median preload is 53 tokens and nine in ten cost 135 tokens or less. The 53-token median is about the length of a short text message.
On-demand skill bodies have a median of 1,302 tokens when loaded. The largest every-session entry listed is bf-replay, at 799 tokens.
At 20 sessions a day, the 53-token median becomes 1,060 tokens. A small preload can matter more than a long body you rarely invoke.
What the test-skill scan can and cannot tell you
The scan graded 13,170 skills A, 300 B, 354 C or D, and 1 E or F. An A means the scan found nothing it treats as risky. It does not mean the skill suits every project.
Finding types include 1,030 network calls, 201 recursive force deletes, 111 root requests, 109 shell commands, and 91 downloads that execute remote code.
The VS Code launch skill is a C-grade example with recursive force delete. gstack qa-only is C with network calls. browser-use qa is C with remote-code downloads. The scan points to the line behind each finding. Read it before installing. It does not decide whether the skill fits your work.
Six testing skills worth starting with
The original is the first copy of a file we saw; everything matching it afterwards is a copy.
- next-cache-components-optimizer: makes a Next.js route load and navigate instantly with an automated end-to-end test; 170 tokens every session, 478 lines, original, 2 copies.
- component-fixtures: creates isolated component setups for repeatable screenshot tests, including services, styles, and themes; 47 tokens every session, 364 lines, original, 1 copy.
- playwright-cli: controls a browser to open pages, fill forms, click controls, take snapshots, and test interactions; 19 tokens every session, 426 lines, original, 29 copies.
- kotlin-testing: guides Kotlin tests with Kotest and MockK, including coroutine and property-based tests; 38 tokens every session, 825 lines, original, 9 copies.
- smoke-test: checks DeerFlow end to end by starting services and checking their health; 0 tokens, waits for a match, 183 lines, original, 2 copies.
- tui-tester: checks Gemini CLI behaviour and terminal appearance through automated interactions; 19 tokens every session, 66 lines, original, 0 copies.
When a hook should run the test instead
A hook is a shell command your agent runs for you when something happens. Choose a hook when a test must run at a fixed event, such as before a push. A skill cannot fire on an event. It can guide the model through a test, but the model must decide to use it first.
Questions people ask
Can an A grade replace reading the instructions? The catalogue records 52 reviewed mods and 0 with high review risk. Treat that as another signal, not permission to skip the relevant lines.
Can a skill run tests by itself? It can tell the model what to run after it is invoked. If the check must happen automatically, choose a hook instead.
How often do these skills change? 4,728 mods changed in the last 7 days. Check the measured date before relying on instructions that call tools or prescribe commands.
Do it now
Run npx agentmods add <slug> to install one, then browse the skills catalogue.