AgenticKanbasutra

Preview VS Code Extension

AgenticKanbasutra

A local command center for turning coding-agent work into planned, queued, executable, and reviewable tasks.

Open in Marketplace View source
AgenticKanbasutra task board running inside a VS Code WebView

Why it exists

Agent prompts should not disappear into chat history.

AgenticKanbasutra keeps each task attached to the repository, branch, runner, permissions, context, output, artifacts, and final review state. It is built for developers who want agent automation without losing visibility or accountability.

  • Manual and automatic queue execution
  • Collapsible repository and runner grouping for busy boards
  • Best-effort provider health updates before queueing work
  • Prompt, stdout, stderr, summary, diff, and changed-file artifacts
  • Failed-task review with quick requeue paths

Inside VS Code

The extension uses a focused WebView, not an external dashboard.

Task board grouped by workflow status
Task board Pending, Queued, Execution, Failed, and Done stay visible, with collapsible groups for dense boards.
New task composer with repository and runner settings
Task composer Define repository, branch, runner, model, permissions, priority, and attached context.
Task detail drawer with summary and artifacts
Review drawer Inspect summaries, changed files, logs, prompts, and failure details after each run.
Provider usage health section in the AgenticKanbasutra sidebar
Provider usage Check Codex and Claude readiness locally, and jump to Copilot usage on the web.
Provider usage warning before queueing a task
Queue warning Review provider limit signals before queueing more work, while keeping override control.

Workflow

From idea to reviewed output.

  1. 01 Create

    Write the spec, attach only useful context, and select the repository, branch, runner, model, tools, permissions, and priority.

  2. 02 Queue

    Keep execution manual, or switch to automatic mode with a configured concurrency limit.

  3. 03 Execute

    Run local provider CLIs, generic commands, or manual handoffs while respecting the selected repository boundary.

  4. 04 Review

    Move successful runs to Done, inspect failures, and requeue only when the cause has been understood.

Runners

Choose the safest execution path for each task.

manual Prompt package plus human-provided output. Best for testing and interactive tools.
generic-cli A configurable local command template for trusted automation.
copilot-cli Local GitHub Copilot CLI attempts with captured output and permission arguments.
claude-cli Local Claude Code CLI execution in non-interactive mode, with explicit Claude model choices.
codex-cli Local Codex CLI execution with sandbox-aware permission mapping.
cloud handoffs Preview dispatch paths for provider-hosted workflows that will mature over time.

Safety

Orchestration does not remove responsibility.

AgenticKanbasutra can launch local commands and provider CLIs that may edit files, create branches, commit, push, or consume paid provider resources. Review what is generated before accepting it.

Task records and run artifacts stay in VS Code extension storage on your machine, but they may contain prompts, notes, logs, paths, summaries, changed files, and diffs.

  • Keep secrets out of specs and attached context
  • Sanitize artifacts before sharing them publicly
  • Prefer worktree isolation for write-capable runs
  • Treat command templates as trusted code
  • Use bypass permissions only when explicitly needed

Preview

Try it with a harmless repository first.

Install preview Open an issue