OpenParable
Skip to content
← Back to skills

Webctl

Apps & Automation
webctl

Webctl is an Agent Skill for apps & automation published in the cosinusalpha/webctl repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Browser automation via CLI. Use when browsing websites, filling forms, extracting data from web pages, taking screenshots, or automating web interactions. Preferred over MCP browser tools for better context control.

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/cosinusalpha/webctl/blob/HEAD/skills/webctl/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/webctl, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

webctl - Browser Automation CLI

Setup

If webctl is not installed, run: pip install webctl && webctl setup

Workflow

# 1. Navigate — returns structured data + page summary
webctl navigate "https://example.com"

# 2. Interact using @refs or text descriptions
webctl click @e3
webctl type "Email" "[email protected]"

# 3. End session
webctl stop

Choosing the Right Approach

GoalCommand
Browse / interact with a pagenavigate URL — returns structured data + page summary (add --snapshot for full a11y with @refs)
Read article contentnavigate URL --read — returns readable markdown
Search a websitenavigate URL --search "query" — types query + returns results
Filter specific elementsnavigate URL --grep "€|price" — filtered a11y snapshot

Commands (10 total)

navigate - Go to URL

webctl navigate "https://example.com"                        # structured data + page summary (default)
webctl navigate "https://example.com" --snapshot              # full a11y snapshot with @refs
webctl navigate "https://example.com" --grep "€|price"       # filtered a11y snapshot
webctl navigate "https://example.com" --read                  # readable text content
webctl navigate "https://duckduckgo.com" --search "query"     # search + results snapshot

The default returns structured data (JSON-LD, Open Graph — price, rating, etc.) plus a page summary. Add --snapshot for the full a11y snapshot with @refs for interaction.

snapshot - Re-scan current page

webctl snapshot                     # all elements with @refs (default)
webctl snapshot --interactive-only  # just buttons/links/inputs
webctl snapshot --read              # readable text content
webctl snapshot --count             # just counts (zero context)
webctl snapshot --grep "pattern"    # filter by regex
webctl snapshot --within "role=main"  # scope to container

click - Click an element

webctl click @e3                    # by @ref (fastest)
webctl click "Submit"               # by text description
webctl click 'role=button name~="Submit"'  # by query (fallback)
webctl click "Submit" --snapshot    # click + return new page state
webctl click "Next" --snapshot --grep "result"  # click + filtered snapshot

type - Type text (auto-detects dropdowns and checkboxes)

webctl type @e2 "[email protected]"  # by @ref
webctl type "Email" "[email protected]" # by text description
webctl type "Country" "Germany"     # auto-selects from dropdown
webctl type "Search" "query" --submit  # type + press Enter
webctl type "Email" "x" --snapshot  # type + return page state

do - Batch multiple actions in one call

webctl do '[["type","Email","[email protected]"],["type","Password","secret"],["click","Log in"]]' --snapshot

Actions: click, type, press, wait. Stops on first failure.

press - Keyboard key

webctl press Enter
webctl press Escape
webctl press Tab

wait - Wait for condition

webctl wait stable                                  # best for SPAs (DOM stabilization)
webctl wait network-idle                            # for traditional page loads (avoid for SPAs with WebSocket/SSE)
webctl wait 'exists:role=button name~="Continue"'
webctl wait 'url-contains:"/dashboard"'
webctl wait 'hidden:role=dialog'

save - Save session state

webctl save                  # save to current session profile
webctl save my-auth          # save as named profile (reusable with -s my-auth)

stop - Close everything

webctl stop                  # closes browser + daemon (default)
webctl stop --keep-daemon    # only close browser

start - Explicit session start (usually not needed)

webctl start                                  # visible browser
webctl start --mode unattended                # headless
webctl navigate URL --mode attended           # auto-start with visible browser

Target Syntax

Actions (click, type) accept three target formats:

FormatExampleWhen to use
@ref@e3After a snapshot (fastest, most reliable)
Text"Submit"When you know the element text
Query'role=button name~="Submit"'When text is ambiguous

@refs are assigned by snapshot and navigate --snapshot/--search/--grep. They reset on each snapshot.

Text descriptions are fuzzy-matched against interactive elements. webctl prefers the right role for the action (click prefers buttons/links, type prefers textboxes).

Query syntax is the fallback: role=X name~="Y" (use name~= for contains, name= for exact).

Automatic Fallbacks

These happen transparently — you don't need to handle them:

  • Cookie/popup auto-dismiss: Overlays are dismissed before actions
  • Scroll-to-find: If element not found, scrolls down and retries (2x)
  • Click retry on overlay: If click intercepted by overlay, dismisses and retries
  • Smart type: type on combobox auto-uses select_option; on checkbox auto-uses check/uncheck

Common Patterns

Price Lookup (e-commerce)

webctl navigate "https://amazon.de/dp/B09HM94VDS"
# Structured data (price, rating) + snapshot with @refs for details
webctl stop

Read Article

webctl navigate "https://spiegel.de" --read
# Returns structured data + readable markdown content
webctl stop

Search and Extract

webctl navigate "https://duckduckgo.com" --search "query"
# Returns search results with @refs
webctl stop

Login

webctl navigate "https://example.com/login"
webctl do '[["type","Email","[email protected]"],["type","Password","secret"],["click","Log in"]]' --snapshot
webctl wait 'url-contains:"/dashboard"'
webctl stop

Form with Dropdown

webctl navigate "https://example.com/form"
webctl do '[["type","Name","John"],["type","Email","[email protected]"],["type","Country","Germany"],["click","Submit"]]' --snapshot
webctl stop

Find Specific Data on Complex Pages

webctl navigate "https://example.com" --grep "€|price|shipping"
# Returns only elements matching the pattern, with @refs
webctl stop

Human-In-The-Loop

For CAPTCHA, MFA, or manual steps (requires visible browser — don't use --mode unattended):

webctl start                                          # visible browser
webctl navigate "https://example.com/login"
webctl type "Email" "[email protected]" --submit
webctl prompt-secret --prompt "Enter MFA code:"       # pauses for human
webctl wait 'url-contains:"/dashboard"'
webctl stop