Wallfacer is a task-board runner for Claude Code. It provides a web-based kanban board where tasks are created as cards, executed by AI agents in isolated sandbox containers, and reviewed when complete.
The guides below are numbered in recommended reading order. Each guide has an Essentials section covering the basics you need to get productive, and an Advanced Topics section for power users. Start from Guide 1 and work forward. You can skip Advanced sections on first read and return to them when you need deeper control.
Install Wallfacer, configure your API credentials, and launch the task board for the first time. Essentials — install, configure, first run.
The core experience: create tasks, run them, and review results on the kanban board.
- Essentials: Creating tasks, running them, reviewing diffs, handling waiting tasks
- Advanced: Batch creation, dependencies, scheduled execution, budgets, soft delete
Where your code lives and how changes flow back to your repositories.
- Essentials: Setting up workspaces, basic git operations (push, sync, branches)
- Advanced: Workspace groups, worktree internals, commit pipeline, conflict resolution, AGENTS.md instructions
Hands-off operation: let Wallfacer promote, test, and submit tasks without manual intervention.
- Essentials: Enabling autopilot, auto-test, auto-submit
- Advanced: Pipeline architecture, auto-retry, circuit breakers, dependency ordering, scheduled execution
AI-assisted prompt improvement and idea generation before tasks are executed.
- Essentials: Refining a prompt, running the ideation agent
- Advanced: Auto-refine, ideation intervals, system prompt customization
Understanding what agents did and what it cost.
- Essentials: Reading oversight summaries, checking costs, live log monitoring
- Advanced: Flamegraph/timeline, span statistics, budget enforcement, stats dashboard
Deep reference for all settings and customization options.
- Essentials: Settings UI walkthrough, key environment variables
- Advanced: System prompt templates, sandbox routing, security, CLI reference
Fault isolation and self-healing automation. Advanced only — circuit breakers automatically pause promotion, testing, and submission when repeated failures are detected, then self-heal via exponential backoff.
Create multiple Backlog tasks, enable Autopilot, and let Wallfacer run them concurrently. Each task works on a separate branch, so there are no conflicts during execution. Conflicts (if any) are resolved at merge time.
- Create a task and run it
- Review the diff and mark it as Done if it looks right, or provide feedback if it needs adjustment
- Continue the feedback loop until the result is satisfactory, then mark Done
- Write a task prompt that includes clear acceptance criteria
- Run the task; when it reaches Waiting, click Test (see Oversight & Analytics)
- If it fails, send feedback with the test output; re-run until passing
- Mark Done to commit
- Enable Autopilot + Auto-Test + Auto-Submit (see Automation)
- Create backlog tasks with dependencies (see Board & Tasks)
- Tasks are automatically promoted, tested, and submitted as they complete
Architecture details, the HTTP API reference, task state machine, container orchestration, and data models are documented in the internals documentation.