[FEATURE] isolated: true flag for slash commands to run with clean context

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

When running multiple slash commands sequentially in one session (e.g., /commandA followed by
/commandB), the second command's execution is polluted by the first command's conversation
history. This causes:

• The main session's orchestration logic being influenced by irrelevant prior context
• Subagents spawned by the command are already isolated, but the main session that orchestrates

them is not

• The only workaround is manually typing /clear between commands, which breaks workflow

automation

Proposed Solution

Add an isolated frontmatter option to slash commands (and skills):

```markdown
---
name: my-command
isolated: true
---

When isolated: true, the command executes as if the conversation history were empty. Only
persistent state (memory, CLAUDE.md, project settings) is retained — prior conversation turns
are excluded from the context window during this command's execution.

Alternative Solutions

• Manual /clear: Works but not automatable, easy to forget
• Wrapping in an Agent: Agents have clean context, but lose the ability to interact with the

user (e.g., confirmation prompts between stages)

• Launching a new claude process via Bash: Loses interactivity entirely

Priority

High - Significant impact on productivity

Feature Category

CLI commands and flags

Use Case Example

Multi-stage workflows where a user runs one command, then immediately runs another command that
orchestrates multiple sub-agents. The orchestration logic should not be influenced by the output
from the previous command.

Additional Context

_No response_