[FEATURE] /workflows support in VSCode extension

@kreare +1. I think reproducibility is going to become a key part of agent development. Being able to replay an identical workflow with prompts, tool calls, MCP interactions, and model responses would make debugging and regression testing much easier. It also helps compare behavior after prompt or model changes.

Yep, /workflows is absolutely essential for the Claude Code VS extension and yet it's not a valid function. The last few months they've been updating the extension often enough that in does not make sense to not include a feature like this. Do they not have an objective to be on parity with Claude CLI? It's so weird this and many other things are just not on parity between the two. Come on Anthropic, you have no excuses. Fix this.

Hi, just want to add some details to this issue as this is something I'm bumping into as well:

Root-cause detail from extension v2.1.269 (bundled binary, build 2026-09-11, Linux x64)

This isn't a missing feature — it's a deliberate dispatch guard, with an interesting contradiction:

1. Typing /workflows in the extension chat panel returns exactly /workflows isn't available in this environment. That string comes from a guard in the bundled CLI (resources/native-binary/claude):

if (l.options.isNonInteractiveSession && mI().has(d)) {
  ... `/${...} isn't available in this environment.`  // telemetry: cmd_unavailable_headless
}

The extension drives the bundled CLI over the Agent SDK control protocol (CLAUDE_CODE_ENTRYPOINT=claude-vscode), so the core marks the session isNonInteractiveSession and refuses all built-in interactive commands there.

1. /workflows is defined as type: "local-jsx" — a terminal React panel ("Browse running and completed workflows") that can't render outside a TUI. A sibling built-in error is explicit: "…opens an interactive panel and isn't available in this environment. Run it from the Claude Code terminal instead."

1. The contradiction: in the same extension session, the Workflow tool itself is fully enabled — workflows run, /deep-research is available, the workflow-authoring skill loads. No account/plan/config gate is involved (no disableWorkflows; the feature defaults on). It's purely the viewer UI that is TUI-only.

Suggested direction: render the workflows viewer in the extension's webview UI, or fall back to a plain-text listing under non-interactive sessions. Today the panel shows in-flight runs only via the task line under the input box; there's no way to browse completed/failed runs or drill into per-agent output — which is what this issue asks for.

Environment: anthropic.claude-code 2.1.269, Ubuntu 22.04, VS Code stable.