Skip to content

[FEATURE] Allow skills to clear context before execution #45091

Description

@dianedef

Problem

When starting a new task via a custom skill (e.g. /my-start-task), you typically want a fresh context window — no leftover tokens from the previous conversation polluting the new session.

Currently, the only way to achieve this is to manually type /clear before invoking the skill. This is a minor but constant friction point in daily workflows, especially for skills designed to be session starters.

Proposed Solution

Allow skills to declare that they need a fresh context. Possible approaches:

Option A — Frontmatter flag

---
name: start-task
clear: true
---

When clear: true is set, Claude Code automatically clears the context window before executing the skill.

Option B — Allow hooks to trigger /clear

Let pre-tool hooks (or a new pre-skill hook type) invoke built-in commands like /clear, so users can wire up their own automation.

Option C — Compound commands

Support chaining built-in commands: /clear && /my-skill arg or /clear | /my-skill arg.

Why This Matters

  • Skills like "start a new task" are natural session boundaries — they should start clean
  • Having to remember /clear before every /start adds friction and is easy to forget
  • Forgetting leads to bloated context, hallucinations from stale conversation, and wasted tokens

Current Workaround

/clear
/my-skill my task

Two commands instead of one, every single time.

Metadata

Metadata

Assignees

No one assigned

    Labels

    duplicateThis issue or pull request already exists

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions