AI commands overview
PACE E2E provides six AI commands. Four are single-agent commands — each runs one agent in isolation, giving you precise control over a single step in the pipeline. The other two are workflow commands (also called pipeline commands) — each chains multiple agents automatically, taking you from input to a reviewed, runnable test in one invocation.
At a glance
| Command | Type | What it does | Start from |
|---|---|---|---|
| /e2e-from-recording | Workflow | Recording → spec → test → review (Stabilizer → Generator → Reviewer) | A VS Code Playwright recording |
| /e2e-from-scratch | Workflow | Scenario → spec → test → review (Planner → Generator → Reviewer) | A plain-English description |
| /plan-test | Single-agent | English → spec file(s) in tests/specs/ | A plain-English description |
| /generate-test | Single-agent | Spec → .spec.ts in tests/e2e/ + smoke run | A finished spec file |
| /review-test | Single-agent | Test (or diff) → PASS / FAIL verdict | A .spec.ts or the latest commit |
| /stabilize-recording | Single-agent | Recording + English → canonical spec + data + alignment report | A VS Code Playwright recording |
Which command should I reach for?
I recorded a flow in VS Code and I want a test file. Use /e2e-from-recording — it's the primary path and does everything in one step.
I want to design a test from scratch without recording anything. Use /e2e-from-scratch — describe the scenario in English and the full pipeline runs automatically.
I want to review or edit the spec before generating TypeScript. Use /plan-test or /stabilize-recording to produce the spec, review it, then run /generate-test manually.
I already have a spec and I just need the .spec.ts.
Use /generate-test directly.
I have a test file and I want to check it against the rulebook. Use /review-test.
Single-agent commands vs workflows
Single-agent commands let you run one step at a time — useful when you want to review the intermediate output (e.g. the spec) before continuing. Workflow commands skip the intermediate checkpoints and run the full pipeline for you.
Both types produce the same underlying files; the only difference is whether the agents are chained automatically or invoked one by one.
None of the commands commit automatically. You always review the output and commit when you are satisfied.
See also
- The four agents — how the Planner, Stabilizer, Generator, and Reviewer work and how they fit together.
- Choose your AI tool — how to pick Claude Code, Codex CLI, Gemini CLI, or Antigravity, and how the per-host tabs work.