Let your coding assistant help choose commands, preview cases or diagnose an error. Start with Run Quick Eval if you just want to try PromptBeat.

Download AIBeat Skill · ZIP

Five instruction packs with their references. No executable or model runtime inside.

Get PromptBeat first

Download the full product package if you have not installed it yet.
ZIP checksum · Legacy archive · Public Skills source The ZIP is the website’s instruction bundle for AIBeat Skill. The linked public repository is an earlier source snapshot; its instructions can differ from this bundle. Full v0.2 product packages also contain an earlier Skills copy. Downloading or loading a Skill does not install PromptBeat or configure your model credentials.

​
1. Extract the Skills ZIP

Use your system’s archive tool, or run one of these commands in your downloads folder. Choose a new destination rather than overwriting an existing copy. 
unzip aibeat-skill.zip -d aibeat-skill-unpacked
Inside aibeat-skill-unpacked/aibeat-skill/ (or promptbeat-skills-unpacked/promptbeat-skills/ for the legacy archive), find promptbeat-run-quick-eval/. Keep its SKILL.md and references/ folder together. MANIFEST.json lists the source files and their SHA-256 checksums.

​
2. Add one Skill to your coding tool

Open the project where you use your coding assistant. Copy the whole promptbeat-run-quick-eval folder to the matching location below. If that folder already exists, compare your local changes before replacing it. You can use your file manager on any platform. 
your-project/
└── .claude/skills/promptbeat-run-quick-eval/
    ├── SKILL.md
    └── references/cli-recipes.md
Start a fresh Claude Code session in that project. Use /promptbeat-run-quick-eval to select the Skill, then provide the task below. Claude Code Skills documentation.
Only the selected Skill is needed for this first task. Install the others the same way when you need them.

​
3. Give the assistant its first task

Replace <package-path> with the extracted PromptBeat product package folder—not the Skills folder. Copy this into your coding assistant: 
Use promptbeat-run-quick-eval. My PromptBeat package is at <package-path>.
Validate examples/bootstrap/promptbeat.yaml and preview up to five local cases
in artifacts/cases.json. Do not call models or change credentials.
Show me where the output is saved.
The expected local steps are: 
mkdir -p artifacts
./bin/promptbeat validate --config examples/bootstrap/promptbeat.yaml
./bin/promptbeat generate --config examples/bootstrap/promptbeat.yaml --count 5 --output artifacts/cases.json
On Windows, the assistant should use .\bin\promptbeat.cmd and PowerShell paths as shown in the quickstart. These commands preview inputs, not model answers or scores. Inspect artifacts/cases.json in the package folder to finish the first task.
The Skill guides your coding assistant; it is not a sandbox or a permission system. Review commands before approving them. For run or eval, configure your providers and explicitly approve the endpoints, scope and possible model costs. Never paste credentials into the conversation.

​
Which Skill should I use next?

Skill directoryUse it to
promptbeat-getting-startedFind your package and choose a first local test.
promptbeat-run-quick-evalChoose between validation, case preview, evaluation and reports.
promptbeat-select-risk-packPick model risks, existing scenarios and local seeds.
promptbeat-debug-runDiagnose config, path, provider and result errors.
promptbeat-connect-coding-agentDistinguish using a coding assistant from testing one, and identify legacy integration limits.
These are AIBeat Skill instruction packs covering PromptBeat workflows, not an AgentBeat installer or evidence collector. For HTTP Agent evaluation, start with AgentBeat setup and release compatibility.

Connect a model and run

Continue from local preview to a full evaluation.

Read the results

Find the HTML report and underlying result files.