TASK GUIDE · SOURCE-AWARE

Best Agent Skills for browser automation

A minimal Skill stack for repeatable browser actions, scope control, and verifiable output.

Build this plan for my Agent ↗
Desired outcome

Convert a repeatable browser workflow into a controlled run with visible success criteria.

Safe working boundary

Use test accounts where possible; do not submit irreversible actions without confirmation.

01 · SKILL STACK

A minimal three-role stack

Each role has one job. The personalized builder can adjust the selection for your Agent without changing the decision logic.

01 · Core execution

agent-browser

Performs the browser navigation and interaction sequence.

Main limitation
The checked-in SKILL.md is a discovery stub; full instructions come from the installed CLI version.
First risk
Browser-side actions
Source evidence
Pinned · 548b159b
Inspect the Skill ↗
02 · Guardrail

minimal-run-and-audit

Limits the first run to the smallest useful path.

Main limitation
It does not choose the reproduction target, train a model, conduct broad paper analysis, or hide scientific-meaning changes.
First risk
Scientific changes hidden as setup fixes
Source evidence
Pinned · 3ab50525
Inspect the Skill ↗
03 · Evidence

playwright-workflows

Checks the final state and preserves repeatable evidence.

Main limitation
Detailed package files are not fully catalogued.
First risk
Review before install
Source evidence
Editorial source
Inspect the Skill ↗

02 · ACCEPTANCE

How to know the task is finished

  1. Run the workflow once with harmless test data.
  2. Verify the final page state and any created record.
  3. Repeat the run from a clean browser state.

03 · STOP CONDITIONS

When to stop

  • The flow reaches payment, publishing, deletion, or account changes.
  • A selector depends on unstable visual position only.
  • The site requests an undeclared credential or permission.

04 · METHOD

How this guide was produced

SkillSignal maps a defined task to three non-overlapping responsibilities, then selects from reviewed catalog metadata. Agent fit, source state, permissions, and published scores are evaluated with deterministic rules. No LLM writes or reorders the recommendation at request time.