Skip to main content

agent-browser

Automate websites with the agent-browser CLI for navigation, form flows, screenshots, extraction, and repeatable browser tasks. Use when users ask to interact with web apps, gather evidence from pages, or run browser-based QA steps.

Source facts

Repository
getsentry/junior
Last source activity
July 21, 2026 at 20:08
Detected SKILL.md language
English
Stars
367
Forks
40

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
11 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
agent-browser
description
Automate websites with the agent-browser CLI for navigation, form flows, screenshots, extraction, and repeatable browser tasks. Use when users ask to interact with web apps, gather evidence from pages, or run browser-based QA steps.
# Browser Automation Use this skill when work requires real browser interaction instead of API-only lookups. ## Core Workflow 1. Open target page: - `agent-browser open <url>` 2. Capture interactive refs: - `agent-browser snapshot -i` 3. Interact by ref: - `agent-browser click @e1` - `agent-browser fill @e2 "value"` 4. Re-snapshot after every navigation or significant DOM change: - `agent-browser snapshot -i` 5. Capture result evidence: - `agent-browser screenshot --annotate <file.png>` ## Operational Rules - Use `agent-browser` directly; do not use `npx agent-browser`. - Prefer one named session for multi-step flows: - `agent-browser --session <name> ...` - Use chained commands with `&&` only when no intermediate output is needed. - For commands that need refs (`@e*`), run `snapshot -i` first and use fresh refs. - If page load is async-heavy, wait explicitly: - `agent-browser wait --load networkidle` - For user-facing findings, include screenshots and the exact page URL. - When a user asks to see/share a screenshot in Slack, use `sendFiles` with the screenshot path in `files` so the image is included in the active conversation. - Never claim a screenshot was shared unless `sendFiles` returned success in this turn. - If `sendFiles` is unavailable or file attachment fails, state the failure clearly and include the saved screenshot path or error instead of claiming success. ## Common Patterns ### Form flow ```bash agent-browser --session demo open https://example.com/signup agent-browser --session demo wait --load networkidle agent-browser --session demo snapshot -i agent-browser --session demo fill @e1 "user@example.com" agent-browser --session demo click @e2 agent-browser --session demo wait --load networkidle agent-browser --session demo screenshot --annotate /tmp/signup-result.png ``` ### Structured extraction ```bash agent-browser open https://example.com agent-browser snapshot --json > /tmp/page.json agent-browser get text body > /tmp/page.txt ``` ## Reference Files Read only what is needed: - Command surface: [references/commands.md](references/commands.md) - Auth vault + login flows: [references/authentication.md](references/authentication.md) - Session/state persistence: [references/session-management.md](references/session-management.md) - Snapshot refs and stale-ref handling: [references/snapshot-refs.md](references/snapshot-refs.md) - Proxy/network setup: [references/proxy-support.md](references/proxy-support.md) - Recording and video capture: [references/video-recording.md](references/video-recording.md) - Performance profiling: [references/profiling.md](references/profiling.md) ## Templates - Authenticated flow bootstrap: [templates/authenticated-session.sh](templates/authenticated-session.sh) - Capture workflow skeleton: [templates/capture-workflow.sh](templates/capture-workflow.sh) - Form automation skeleton: [templates/form-automation.sh](templates/form-automation.sh)
View on GitHub