24 public agents from 9 publishers, placed on the delivery lifecycle. These run their own loop — hand one a job and it keeps going without you.
The registry calls everything a skill. That flattens the distinction that matters most when you are deciding what to install. A skill is knowledge the model reads and applies to whatever it was already doing — it loads, it shifts the output, you move on. An agent takes the wheel: it spawns subagents, drives a browser, writes artifacts, and runs a multi-step loop of its own.
That difference is a budget. An agent costs tokens and wall-clock time, and it needs a task actually worth handing over — a test suite to generate, a codebase to audit, a browser flow to exercise. Installing one you never invoke costs you nothing but context; invoking one casually costs real money. The 59 entries on the skills page are the cheap half.
Ordered by delivery phase. Anything that spans two phases is listed under both — the phase chips on each card say which.
Turning an intention into something an agent can execute without inventing the requirements as it goes.
Interviews you about your own plan, relentlessly, until the assumptions that were never stated are on the table.
Reach for it when You are about to start building and the plan feels fine — which is exactly when it has not been tested.
npx skills add mattpocock/skillsInstalls all of skills, not this one alone.
Turns a conversation into a product requirements document somebody else could pick up.
Reach for it when The decision happened in a thread and now needs to exist somewhere a new joiner can read it.
npx skills add mattpocock/skillsInstalls all of skills, not this one alone.
Interfaces, module boundaries and domain language — decided before the code sets them in concrete.
Produces a threat model for a feature or system — assets, entry points, trust boundaries, mitigations.
Reach for it when A new externally-reachable surface, before it is built rather than after it is pentested.
npx skills add openai/skillsInstalls all of skills, not this one alone.
Writing the change: framework-specific conventions, and the discipline that keeps a long agent run on the rails.
Works through a written plan step by step, checking off as it goes and stopping when a step does not hold.
Reach for it when You have a plan and want it executed in order, not reinterpreted halfway through.
npx skills add obra/superpowersInstalls all of superpowers, not this one alone.
Splits work across subagents with their own context windows so a long task does not degrade as the transcript grows.
Reach for it when The job is big enough that quality visibly drops before it finishes.
npx skills add obra/superpowersInstalls all of superpowers, not this one alone.
Runs genuinely independent work concurrently, with the rules for when that is a speedup and when it is a merge conflict.
Reach for it when Several unrelated changes across a repo that do not touch the same files.
npx skills add obra/superpowersInstalls all of superpowers, not this one alone.
Builds a Model Context Protocol server — tool design, transport, schema, the mistakes that make a server unusable.
Reach for it when You are exposing an internal system to agents and want the tool surface designed rather than generated.
npx skills add anthropics/skillsInstalls all of skills, not this one alone.
Proving the change does what it claims — and finding the assertions that were never written.
Drives a real browser against a running app and reports what it found.
Reach for it when You want the change exercised end to end rather than asserted about in a unit test.
npx skills add anthropics/skillsInstalls all of skills, not this one alone.
Uses the DevTools protocol to inspect, measure and reproduce behaviour in a live page.
Reach for it when A bug that only exists in the browser, and only with the real network and real data.
npx skills add addyosmani/agent-skillsInstalls all of agent-skills, not this one alone.
Finds the behaviour that no test currently exercises and reports it as a list.
Reach for it when Inheriting a service with a suite nobody trusts and no idea where the holes are.
npx skills add dotnet/skillsInstalls all of skills, not this one alone.
A multi-agent pipeline that researches, plans and then generates a working test suite — language-agnostic despite the publisher.
Reach for it when A module with no tests at all, where the first hour is deciding what is even worth asserting.
npx skills add dotnet/skillsInstalls all of skills, not this one alone.
Reading a diff critically, and the far harder skill of acting on what a reviewer said.
Dispatches a reviewer with fresh context, so the review is not done by the model that just wrote the code.
Reach for it when Before you open the PR — a self-review from the authoring context finds almost nothing.
npx skills add obra/superpowersInstalls all of superpowers, not this one alone.
Security-focused review of a PR, commit or diff — risk-first, evidence-backed, with stated coverage limits.
Reach for it when A change touching auth, crypto, value transfer or an external call, where a generic reviewer will not go deep enough.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Checks that a fix actually closes the reported issue and did not just move it.
Reach for it when Verifying a security patch before you tell the reporter it is resolved.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Works through review comments on a pull request and responds to or resolves each one.
Reach for it when A PR with twenty review threads, most of which are small and unambiguous.
npx skills add openai/skillsInstalls all of skills, not this one alone.
Threat modelling, scanning, supply chain, and finding the other five copies of the bug you just found.
Security-focused review of a PR, commit or diff — risk-first, evidence-backed, with stated coverage limits.
Reach for it when A change touching auth, crypto, value transfer or an external call, where a generic reviewer will not go deep enough.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Checks that a fix actually closes the reported issue and did not just move it.
Reach for it when Verifying a security patch before you tell the reporter it is resolved.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Produces a threat model for a feature or system — assets, entry points, trust boundaries, mitigations.
Reach for it when A new externally-reachable surface, before it is built rather than after it is pentested.
npx skills add openai/skillsInstalls all of skills, not this one alone.
Finds the other instances of a bug you already found — one root cause usually has several manifestations.
Reach for it when The moment a vulnerability is confirmed. Fixing only the reported instance is the most common incomplete fix there is.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Audits dependencies and their provenance for supply-chain risk.
Reach for it when A release with new third-party dependencies, or a periodic sweep of what you already ship.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Audits CI workflows that invoke AI agents — where prompt injection becomes a repository write.
Reach for it when You just gave an agent a token in CI. This is the review of that decision.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Branches, pipelines, releases and the deploy itself.
Audits dependencies and their provenance for supply-chain risk.
Reach for it when A release with new third-party dependencies, or a periodic sweep of what you already ship.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Reads a failing CI run, reproduces the failure and pushes the fix.
Reach for it when A red pipeline on a PR you already understand, where the loop is mechanical.
npx skills add openai/skillsInstalls all of skills, not this one alone.
Works through review comments on a pull request and responds to or resolves each one.
Reach for it when A PR with twenty review threads, most of which are small and unambiguous.
npx skills add openai/skillsInstalls all of skills, not this one alone.
Takes a project from local to a live Vercel deployment.
Reach for it when First deploy of a new app, or wiring preview deploys into a workflow.
npx skills add vercel-labs/agent-skillsInstalls all of agent-skills, not this one alone.
Triages issues and pull requests into something a maintainer can work through.
Reach for it when An issue tracker that has stopped being read because it is too long.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
Production reality: debugging, performance, instrumentation, and writing down why the system is the way it is.
Uses the DevTools protocol to inspect, measure and reproduce behaviour in a live page.
Reach for it when A bug that only exists in the browser, and only with the real network and real data.
npx skills add addyosmani/agent-skillsInstalls all of agent-skills, not this one alone.
Triages issues and pull requests into something a maintainer can work through.
Reach for it when An issue tracker that has stopped being read because it is too long.
npx skills add trailofbits/skillsInstalls all of skills, not this one alone.
A diagnosis loop that ends in a regression test, not just a green build.
Reach for it when A production bug where the fix matters less than making sure it cannot come back.
npx skills add mattpocock/skillsInstalls all of skills, not this one alone.
Finding skills, writing your own, and keeping a skill library from turning into a junk drawer.
Searches the registry for a skill that fits what you are doing, and installs it.
Reach for it when The honest answer to 'is there already a skill for this' — there are 20,000 of them.
npx skills add vercel-labs/skillsInstalls all of skills, not this one alone.
Anthropic's own tool for writing a new skill in the format agents actually load.
Reach for it when You have internal knowledge worth encoding and want it to match the house format.
npx skills add anthropics/skillsInstalls all of skills, not this one alone.
Same CLI as skills — there is no separate agent installer, because the registry does not distinguish them. The split on this site is ours:
$ npx skills add obra/superpowers
An agent you install is an agent your model can invoke. The unit is the repository, so one command can add several autonomous loops at once. That is fine — they only cost you when something calls them — but it is worth knowing what arrived. Read the entry before you hand it a job, not after.
Agents fail differently from skills. A skill that misses the mark leaves you with a slightly worse diff; an agent that misses the mark can spend twenty minutes and a lot of tokens producing confident, wrong work. Three things worth doing every time:
Nothing on this page is ours. Redline does not vendor, fork, host or modify any of these. Each entry links to the publisher's repository, and the install command pulls from them directly. If a publisher changes an agent, you get their change — this is a map, not a mirror.
The skill/agent split is a judgement we made, not a field the registry publishes. An entry is an agent here if it runs a multi-step loop of its own rather than informing one you are already running. Reasonable people would draw a couple of these differently.
An agent writing code is still code arriving in a pull request, and nothing on this page reviews it. That is the half Redline does — see the output contract and the merge gate. The more autonomous the thing writing the diff, the more the gate is the only thing standing between it and main.