---
title: "Agent skills | Speedscale Docs"
description: "The agent skills Speedscale publishes for AI coding agents, what each one is for, and a prompt that installs them all."
canonical: "https://docs.speedscale.com/proxymock/agent-skills/"
---

[Documentation index](https://docs.speedscale.com/proxymock/llms.txt) | [Markdown](https://docs.speedscale.com/proxymock/agent-skills/index.md) | [Canonical page](https://docs.speedscale.com/proxymock/agent-skills/)

# Agent skills

Speedscale publishes agent skills that teach an AI coding agent (Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Kiro, or any agent that can read a URL) how to install, run, tune, and test with Speedscale and proxymock.

Every skill ships inside proxymock and is mirrored to the [speedscale/skills](https://github.com/speedscale/skills) repository on each proxymock release, so both always match. This page lists what each one is for. Follow a skill's link for its full instructions.

The local skills need [proxymock](https://docs.speedscale.com/proxymock/getting-started/installation/) installed. Cloud replays also need a Speedscale account.

[Watch automatic testing with proxymock MCP skills on YouTube](https://www.youtube.com/watch?v=2SW9cLmQU-U).

<a id="install-the-skills"></a>

## Install the skills

Paste this prompt into your coding agent:

```text
Install the Speedscale agent skills from https://github.com/speedscale/skills. If you are Claude Code, run `/plugin marketplace add speedscale/skills` and then `/plugin install speedscale@speedscale-skills`. Otherwise run `npx skills add speedscale/skills`. If neither works, download every folder under skills/ in that repo into your skills directory. Then list the skills you installed and what each is for.
```

If you already have proxymock, select a client explicitly: `proxymock mcp install --client claude-code --skills` installs MCP and bundled skills for that client in the current project. Use `--skills-only` to omit MCP, `--scope user` for a global installation, and `--dry-run` to preview paths. Managed installs preserve edited or unowned files. Inspect project skills with `proxymock mcp skills list --dir ./.claude/skills`. The separate `proxymock mcp skills export --dir <directory>` command writes to an explicit directory and replaces existing files.

If you use the ChatGPT desktop app with Codex, follow the [ChatGPT section of the skills README](https://github.com/speedscale/skills#chatgpt-desktop-app).

<a id="prompts-in-the-dashboard"></a>

## Prompts in the dashboard

The Speedscale dashboard offers ready-made prompts wherever a task starts. Each one names the skill to use, carries the IDs and your Speedscale host, and tells the agent what to ask you before it acts. Paste it into your agent from your project directory. A prompt never contains your API key or other secrets.

| Where | What the prompt does |
| --- | --- |
| Home | Installs Speedscale with `install-speedscale`. |
| Add Service dialog | Adds a service with `record-traffic`. Before you pick a workload, the agent asks which service and where it runs, and the prompt names your connected clusters. After you pick a workload, the prompt names its cluster, namespace, and workload. The agent asks before it turns on capture or restarts a workload, and it leaves cluster capture on when it finishes, so the service keeps sending traffic. It then tells you how to turn capture off. |
| Add Service, Kubernetes, No Operator Installed | Installs the operator with `install-speedscale`, as an alternative to the Helm command. |
| Add Service, Kubernetes, Record Locally | Records your service on your machine with `record-traffic`. The recording stays in your project directory. |
| Snapshot and report pages | Ranks the prompts that fit the snapshot or report, such as analyzing a report, tuning the tests, or turning a snapshot into a regression test. Open them from the page or from **More actions** > **Use with coding agent**. |
| Replay launcher and replay wizard | **Copy agent prompt** next to the command runs exactly the command shown with `run-snapshot-replay` and follows it to a result. It is unavailable when the command cannot express the form (for example JWT resigning, an edited test config that has not been saved, or a debug run) and when a target URI includes credentials. |

The prompts start in **Fetch skill** style, which tells the agent to download the skill first and works in any agent. If you already installed the skills, switch to **Skill installed** for a shorter prompt. The dashboard remembers your choice in this browser.

<a id="set-up"></a>

## Set up

| Skill | Use it to |
| --- | --- |
| [install-speedscale](https://github.com/speedscale/skills/tree/main/skills/install-speedscale) | Install the `speedctl` and `proxymock` CLIs and the Speedscale Operator, set up your API key, verify the install, and connect the proxymock MCP server to your agent. It also handles upgrades, repairs, and uninstalls. |

<a id="run-and-tune-replays"></a>

## Run and tune replays

These skills work with Speedscale cloud snapshots and with local proxymock recordings.

| Skill | Use it to |
| --- | --- |
| [run-snapshot-replay](https://github.com/speedscale/skills/tree/main/skills/run-snapshot-replay) | Run a snapshot or recording as a replay where it was recorded, on your machine, in a cluster through your kubeconfig, or through Speedscale cloud, and follow it to a result. In a cluster it runs as a regression check or a load test with a test config's goals. |
| [tune-snapshot-replay](https://github.com/speedscale/skills/tree/main/skills/tune-snapshot-replay) | Tune the tests: loop on a replay until its responses are accurate, changing one thing per run (test config assertions, volatile fields, ids created during the session) and keeping or reverting it. Mock problems go to `improve-mock-match-rate`. |
| [analyze-replay-report](https://github.com/speedscale/skills/tree/main/skills/analyze-replay-report) | Explain a cloud report or a local replay run: why it passed or failed, the first failing response, and what to do next. |
| [improve-mock-match-rate](https://github.com/speedscale/skills/tree/main/skills/improve-mock-match-rate) | Tune the mocks for any technology (HTTP, SQL, Redis, Kafka, gRPC, and so on). It applies match-rate fixes offline, can re-run the snapshot against the same workload, compares recorded with mocked traffic, and fixes each discrepancy by its kind. |

<a id="test-with-recorded-traffic"></a>

## Test with recorded traffic

These skills form the proxymock quality loop. They run locally and need no Speedscale Cloud account. Start with `quality-loop` if you are not sure which one applies.

| Skill | Use it to |
| --- | --- |
| [quality-loop](https://github.com/speedscale/skills/tree/main/skills/quality-loop) | Pick the right skill or proxymock command for a task, take a service from no recording to a first regression gate, and check that your environment is ready. |
| [record-traffic](https://github.com/speedscale/skills/tree/main/skills/record-traffic) | Record a service's inbound and outbound traffic, including its databases, by wrapping it with `proxymock record`, or record a Kubernetes workload with eBPF capture and pull the traffic into the workspace, then confirm every dependency was captured. |
| [proxymock-regression-test](https://github.com/speedscale/skills/tree/main/skills/proxymock-regression-test) | Replay a recording at your service and fail on status or body changes against a known-good baseline. A cluster workload goes to `run-snapshot-replay` in its regression mode. |
| [proxymock-verify-fix](https://github.com/speedscale/skills/tree/main/skills/proxymock-verify-fix) | Prove a bug fix by replaying the incident traffic at the fixed build. |
| [proxymock-contract-test](https://github.com/speedscale/skills/tree/main/skills/proxymock-contract-test) | Check traffic against an OpenAPI spec, or mock a dependency from its spec before you have a recording. |
| [proxymock-chaos-mock](https://github.com/speedscale/skills/tree/main/skills/proxymock-chaos-mock) | Make a mocked dependency return errors, slow responses, corrupt bodies, or connection faults to test retries and timeouts. |
| [proxymock-load-test](https://github.com/speedscale/skills/tree/main/skills/proxymock-load-test) | Replay recorded traffic with parallel virtual users and report latency percentiles, throughput, and match rate. A cluster workload goes to `run-snapshot-replay` in its load mode. |
| [proxymock-perf-container](https://github.com/speedscale/skills/tree/main/skills/proxymock-perf-container) | Load-test one service with its dependencies mocked, and tell an app limit from a test harness limit. |
| [proxymock-compare-results](https://github.com/speedscale/skills/tree/main/skills/proxymock-compare-results) | Compare two replays or recordings and show what regressed, improved, or stayed the same. |
| [proxymock-summarize-recording](https://github.com/speedscale/skills/tree/main/skills/proxymock-summarize-recording) | Summarize a recording: hosts, endpoints, methods, status codes, and volume. |

<a id="removed-skills"></a>

## Removed skills

*   `proxymock-replay-tuning` was folded into `improve-mock-match-rate`. If your agent still lists it, install the skills again to pick up the current set.

<a id="related"></a>

## Related

`install-speedscale` can connect the proxymock MCP server to your agent. It asks which client and scope you want, previews the paths, and installs only that selection. See [Model Context Protocol (MCP)](https://docs.speedscale.com/proxymock/how-it-works/mcp/) for how that server works.
