Skill Detail

Record an agent-completed browser task as a parameterized Oya playbook

Complete a test browser task once, save its learned steps as an Oya playbook, and verify a replay with new inputs while retaining repair evidence.

Browser AutomationMulti-Framework
Browser Automation Multi-Framework Security Reviewed
โญ 349 GitHub stars โฌ‡ 4.5k/wk npm
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill record-agent-browser-task-as-parameterized-oya-playbook Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
At a glance
Tools required
Node.js 20+, @oya-ai/browser, configured Oya control plane or cloud API key, model API route; Docker/Compose for self-hosting
Install & setup
Run npm install @oya-ai/browser@1.0.233 in a Node.js 20+ project. For an existing self-hosted deployment set OYA_BASE_URL and OYA_API_KEY; for Oya Cloud follow upstream account setup and set OYA_API_KEY. Configure the model route for learning, then adapt the pinned examples/sdk/05-playbooks.ts sample using a test form and clean up the browser after each run.
Author
OyadotAI
Publisher
Community
Last updated
Nov 12, 2026
Quick brief

Invoke this workflow when an agent has a repeated browser form task and should learn the steps once, then replay them with new inputs. Supply a test URL, a bounded task, named input variables, expected page results, and an authorized Oya browser connection. The completion artifact is a named playbook plus evidence that a second input set produced the expected result.

How it works

What this skill actually does

Use Node.js 20 or later and install the exact-source package with npm install @oya-ai/browser@1.0.233. Connect to an existing self-hosted control plane with OYA_BASE_URL and OYA_API_KEY, or use the documented Oya Cloud setup. Self-hosting additionally requires Docker and Compose. A model API route is needed for learning and any enabled repair. The package’s registry repository points to this upstream’s packages/sdk directory.

Following the upstream playbook example, create an Oya client, start a browser with oya.browser.start(), and navigate with browser.goto(). Call browser.ask() with the bounded task and a data object whose named placeholders represent the changing inputs. Use a test form for the first run. Inspect the completed page before saving the successful run with browser.toPlaybook(). Retain its name, step count, variable names, and exported Playwright code.

Call browser.play() with the saved name and a different input set. Verify the resulting page state and record the replay result. Healthy recorded paths with fixed input fields avoid model calls. Generated free-text answer fields can still invoke the model; provide fixed field values when deterministic replay is required. If automatic repair is enabled and the page changes, inspect the healed result and any revised playbook or draft before reuse. For strict replay, use the documented autoHeal setting to stop on mismatch. Stop the browser in cleanup, including error paths.

Return the playbook identifier, input schema, exported code, observed results for both input sets, and repair or failure details. The scope excludes general browsing and healthcare compliance claims. The community repository does not itself provide the cloud or enterprise BAA offering. Multi-Framework is appropriate because this bounded SDK procedure can run from agent runtimes that execute Node.js; the desktop MCP connection is optional. Reviewed release example. No installation, live replay, or security audit was performed during intake.