[FEATURE] URL parameters for Claude Code on the Web - pre-fill prompts with user confirmation

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I want to integrate Claude Code on the Web into my existing workflows and tools, but there's currently no way to programmatically start a coding session with a pre-defined prompt or task.My workflow involves:

Maintaining a dashboard/tool that tracks common development tasks
Using bookmarklets or scripts to quickly initiate specific types of code reviews or fixes
Sharing standardized task templates with my team (e.g., "Review PR for security issues in [repo]")
Automating the handoff from issue tracking systems to coding assistants
Currently, I must:

Manually navigate to claude.ai/code
Select my repository
Type out the same prompt I've already written elsewhere
Start the task
This breaks the flow and creates friction when I'm trying to rapidly iterate or when team members need to execute standardized tasks. I want to eliminate these manual steps while maintaining security and user control.

Proposed Solution

Add URL parameter support to Claude Code on the Web that allows pre-filling (but not auto-executing) prompts. The ideal flow would be:
User clicks a link like: https://claude.ai/code?repo=user/myrepo&prompt=Fix+the+authentication+bug+in+auth.py

Claude Code on the Web opens with:

The repository pre-selected (if the user has access)
The prompt pre-filled in the input field
A clear visual indicator that this was loaded from a URL
The user must explicitly click "Start task" to begin execution

Optional confirmation dialog for first-time URL parameter usage explaining that:

Links can pre-fill prompts but never auto-execute
The user should review prompts before starting tasks
Option to "Don't show again" after acknowledgment

Key security principles:

URL parameters only pre-fill content, never trigger execution
User always has final approval before Claude starts working
Clear visual indication when content came from a URL
Works similar to mailto: links (pre-fills email but user sends)
Suggested URL parameters:

prompt - The task description
repo - Repository selector (user/repo format)
branch - Optional branch to work on
env - Optional environment name

Alternative Solutions

Current workarounds I've considered:

Copy-paste from external tools - Manual and error-prone
Custom slash commands - Requires setup in each repo, doesn't help with sharing templates externally
Bookmarklets with clipboard injection - Fragile and requires multiple steps
Terminal-based claude CLI with & prefix - Only works from terminal, not from web-based tools or dashboards

Why these don't fully solve the problem:

They all require additional manual steps or local setup
Can't easily share task templates with non-technical team members via links
Don't integrate well with web-based project management or documentation tools

Note about the old claude.ai/new?q= parameter:
This used to exist for regular Claude chat but appears to have been removed (as of October 2025 per GitHub issue #8827). If this was removed for security reasons, the same concerns should inform this feature's design.

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

Primary Use Case: Linear Task Viewer Integration
I've built LinearViewer, a UI for Linear that uses AI to automatically generate coding prompts from Linear tasks. The current workflow has a frustrating "last mile" problem:
Current workflow:

Open LinearViewer and browse Linear tasks
AI generates a detailed, context-aware prompt for implementing the task
Manually copy the generated prompt
Navigate to claude.ai/code in a new tab
Select the repository
Paste the prompt
Click start

With URL parameters, this becomes:

Open LinearViewer and browse Linear tasks
AI generates a detailed, context-aware prompt
Click "Start with Claude Code" button
Claude Code on the Web opens with prompt pre-filled and repository selected
Review and click "Start task"

Example implementation in LinearViewer:
javascript// Generate the Claude Code URL with the AI-generated prompt
const claudeCodeUrl = new URL('https://claude.ai/code');
claudeCodeUrl.searchParams.set('repo', issue.repository); // e.g., 'mycompany/api'
claudeCodeUrl.searchParams.set('prompt', aiGeneratedPrompt);
claudeCodeUrl.searchParams.set('branch', feature/${issue.identifier.toLowerCase()});

// "Start with Claude Code" button
<button onClick={() => window.open(claudeCodeUrl, '_blank')}>
🤖 Start with Claude Code
</button>
Concrete example:

Linear task: "LIN-123: Add rate limiting to authentication endpoint"
AI-generated prompt: "Implement rate limiting for the /api/auth/login endpoint. Requirements: max 5 attempts per minute per IP, return 429 status with Retry-After header, store attempt counts in Redis with 60s TTL, add unit tests for rate limit logic."
Button opens: https://claude.ai/code?repo=mycompany/api&prompt=Implement+rate+limiting...&branch=feature/lin-123
Developer reviews the pre-filled prompt and starts the task

Additional scenarios:
Scenario 2: Internal Documentation "Quick Fix" Links
Our team wiki has troubleshooting guides with embedded fix commands:
markdown### Database Migration Failed?
Common causes: schema conflicts, missing foreign keys

🤖 Start automated diagnosis
Scenario 3: Code Review Bookmarklet
I've created a browser bookmarklet that extracts the current GitHub PR context:
javascriptconst pr = getPRContext(); // Gets PR #, title, files changed
const url = https://claude.ai/code?repo=${pr.repo}&prompt=Review PR #${pr.number}: ${pr.title}. Focus on: security vulnerabilities, performance issues, test coverage.;
window.open(url);

**Scenario 4: Slack Bot Integration**
Team members can request code assistance in Slack:

/claude-task Fix the memory leak in user-service
→ Bot responds with: "Start this task: [Open in Claude Code]"
→ Link opens Claude Code with full context from the Slack thread

Additional Context

Similar features in other tools:

mailto:subject=...&body=... - Pre-fills email clients but user sends
VSCode's vscode:// URLs - Can open files/projects programmatically
GitHub's https://github.com/user/repo/compare/branch - Deep linking with context
Figma's share links with ?node-id= parameters

Security considerations:

Primary risk: Malicious actors could craft links that trick users into executing harmful tasks
Mitigation: Never auto-execute; always require explicit user confirmation
Consider: Visual warning for first-time URL parameter usage
Consider: Domain allowlist for auto-repo-selection (only allow repos user has access to)
Consider: Character/length limits on URL parameters

Technical notes:

URL encoding would handle special characters in prompts
Could implement as progressive enhancement (degrades gracefully if parameters not supported)
Consider rate limiting or CSRF tokens if concerned about automated abuse