Claude Code Web

Running tasks on Claude Code Web with sandboxed branch access

Claude Code Web is a browser-based Claude Code environment that operates with restricted git permissions — it can only push to its own implementation branch. It cannot access aitask-locks, aitask-data, or main branches directly. This guide covers the end-to-end workflow for running aitasks on Claude Code Web.

Standard vs Claude Web Workflow

Step	Standard (local)	Claude Code Web
1. Lock task	Automatic (during `/aitask-pick`)	Manual pre-lock from local machine
2. Implement	`/aitask-pick` — full branch access	`/aitask-pickweb` — current branch only
3. Archive	Automatic (during `/aitask-pick`)	`/aitask-web-merge` — run locally after

Standard workflow handles everything in one session: lock → implement → commit → archive → push.

Claude Web workflow splits into three stages across two environments:

Local machine          Claude Code Web           Local machine
─────────────          ───────────────           ─────────────
1. ait lock 42         2. /aitask-pickweb 42     3. /aitask-web-merge
   (lock task)            (implement + commit)      (merge + archive)

Step-by-Step Guide

Step 1: Pre-Lock the Task (Local)

Before starting a Claude Web session, lock the task from your local machine to prevent another developer or agent from picking it concurrently.

Option A — Using the board TUI:

ait board

Open the task details and click the Lock button (🔒). The board prompts for your email and acquires the lock.

Option B — Using the CLI:

ait lock <task_id>

Pre-locking is recommended but not required. If you skip it, /aitask-pickweb will display an informational warning if the task is locked by someone else, but will proceed regardless.

Step 2: Implement on Claude Code Web

Start a Claude Code Web session on a branch for the task, then run:

/aitask-pickweb <task_id>

For child tasks, use the parent_child format:

/aitask-pickweb 42_2

The skill handles everything autonomously with zero interactive prompts (except plan approval via ExitPlanMode):

Loads the remote execution profile
Resolves and reads the task file
Performs a read-only lock check (informational only)
Creates or verifies the implementation plan
Implements the approved plan
Auto-commits all changes
Writes a completion marker to .aitask-data-updated/

See /aitask-pickweb for the full skill reference.

Step 3: Merge and Archive (Local)

After the Claude Web session completes and the branch is pushed, run locally:

/aitask-web-merge

This skill:

Scans remote branches for completion markers (.aitask-data-updated/completed_t*.json)
Merges the implementation branch to main (excluding .aitask-data-updated/)
Copies the plan file to aiplans/ on the aitask-data branch
Archives the task (and parent if all children are complete)
Pushes main and aitask-data
Deletes the remote implementation branch

If multiple completed branches exist, the skill offers to process them sequentially.

See /aitask-web-merge for the full skill reference.

The `.aitask-data-updated/` Directory

Since Claude Web cannot write to the aitask-data branch, /aitask-pickweb stores task metadata in a .aitask-data-updated/ directory on the implementation branch. This directory contains:

File	Purpose
`plan_t<task_id>.md`	Implementation plan (normally stored in `aiplans/`)
`completed_t<task_id>.json`	Completion marker — signals to `/aitask-web-merge` that the branch is ready

The completion marker JSON includes:

{
  "task_id": "42",
  "task_file": "aitasks/t42_implement_auth.md",
  "plan_file": ".aitask-data-updated/plan_t42.md",
  "is_child": false,
  "parent_id": null,
  "issue_type": "feature",
  "completed_at": "2026-02-24 15:30",
  "branch": "claude-web/t42"
}

During /aitask-web-merge, the plan file is copied to aiplans/ and the entire .aitask-data-updated/ directory is excluded from the merge to main via git rm -rf.

Execution Profile

/aitask-pickweb requires an execution profile from aitasks/metadata/profiles/. It auto-selects a profile named remote if one exists, falling back to the first available profile alphabetically.

The default remote.yaml profile works for both /aitask-pickrem and /aitask-pickweb — pickweb simply ignores fields it doesn’t recognize (lock, archival, review settings). Only these fields are used by pickweb:

Key	Default	Description
`plan_preference`	`use_current`	How to handle existing plans: `use_current`, `verify`, or `create_new`
`post_plan_action`	`start_implementation`	Action after plan approval

See /aitask-pickweb Execution Profiles for details.

Troubleshooting

Branch not detected by `/aitask-web-merge`

Ensure the Claude Web session pushed the branch to the remote
Run git fetch --all and check if the branch appears in git branch -r
Verify the branch contains .aitask-data-updated/completed_t*.json

Merge conflicts during `/aitask-web-merge`

The skill uses git merge --no-ff --no-commit and will prompt you to resolve conflicts interactively. You can also abort the merge and handle it manually.

Task already locked by someone else

/aitask-pickweb performs a read-only lock check — it warns but always proceeds. If you want to prevent concurrent work, pre-lock the task from your local machine before starting the Web session.

“Web workflow requires an execution profile”

Create a profile at aitasks/metadata/profiles/remote.yaml. The default ait setup creates one automatically. Minimal content:

name: remote
description: Fully autonomous workflow - no interactive prompts
plan_preference: use_current
post_plan_action: start_implementation