When a pull request turns green, the question is which mod belongs in the release step. A skill is a folder of instructions the model opens when it decides it needs them. 14,709 of these skills load a small description in every session, so even the first choice affects your work. Timing matters.
What the release shelf holds
The context is everything the model can see while it answers you. Choose a skill for the job you need, not for its repository popularity.
Continuous integration, or CI, automatically builds and tests code after changes. playwright-test-results queries stored CI results for flaky, slow and failed tests. deploy-release-test validates a specific preview package before running a full deployment test.
Cloud work needs a clear destination. vercel-deploy publishes an application and returns preview and claim links. land-and-deploy covers merging, deployment and checking the live service afterwards.
The small preload before deployment starts
A token is the unit your AI plan bills by, about three quarters of a word. The median preload is 57 tokens every session. Across 20 sessions in a day, that median becomes 1,140 tokens. On this shelf, one skill in ten costs 142 tokens or more.
The largest preload is 969 tokens in open-forge, across 312 lines. That is roughly a full screen of editor text. Among 15,154 skills, 445 wait to be asked for instead of loading every session.
What the infrastructure scan flags
The scan grades 13,404 skills A, 900 B, 849 C or D, and 1 E or F. It found 2,611 network-call findings, 773 requests for root access, 381 recursive force deletes and 298 cases that download and execute remote code.
The here-now skill is graded C for subtle steering, root access and network calls. azure-pipelines and packaging-static-installer are graded D for combinations including root access, remote code or force deletion.
Each finding shows the line that triggered it. Read that line yourself. The scan flags behaviour but does not decide whether the workflow belongs in your project. An A can still carry a low-risk capability note, as land-and-deploy does for network calls.
Six starting points for deployment and CI
The original is the first copy of a file we saw; everything matching it afterwards is a copy. All six picks below are original files. Their tails show tokens paid every session, file lines and copies where recorded.
- playwright-test-results: queries a DuckDB database for CI failures, flaky tests and slow tests. 45 tokens every session, 211 lines.
- deploy-release-test: validates a Next.js preview package before running the complete deployment test suite. 115 tokens every session, 130 lines.
- vercel-deploy: publishes an application or website to Vercel and returns deployment links. 69 tokens every session, 113 lines, 9 copies.
- land-and-deploy: merges a pull request, waits for checks, deploys the change and checks service health. 14 tokens every session, 1,046 lines.
- github-release-management: coordinates versioning, testing, builds, deployments and rollbacks for a release. 27 tokens every session, 1,082 lines, 12 copies.
- pinggy-tunnel: gives a local service a temporary public address through an SSH tunnel. 15 tokens every session, 310 lines, 7 copies.
When a different trigger is the right tool
A glob is a filename pattern, like *.tsx. A rule with a glob fits when the instruction belongs only beside matching files.
A hook is a shell command your agent runs for you when something happens. Use one when a push, test completion or deployment event must start an action. A skill cannot be made to fire on an event, though it can tell the model to run a script. Choose the skills hub for work the model should perform when asked.
Questions people ask
Will these skills work with Cursor? Of these, 87 are written for Cursor, 14,466 for Claude Code and 7,310 for Codex, and a skill can be labelled for more than one. Check the target agent before choosing.
What should make me stop before installing one? Treat root access and remote-code execution as review thresholds. That is 849 of the 15,154 skills here at C or D, and 1 at E or F, so read the finding line before allowing either into a deployment workflow.
Should I install a whole plugin for one cloud task? Only if you need most of its bundled skills. Otherwise, choose one narrow catalogue entry for the task you actually have.
Do it now
Run npx agentmods add <slug> after choosing from the skills hub.