Update Context

Extract information learned during conversation and persist it to context files for future sessions.

Arguments

/update-context [filename] [instructions]

• filename (optional): Target file to update. Defaults to CLAUDE.md in project root. Common targets: CLAUDE.md, agents.md, .claude/settings.md
• instructions (optional): Specific guidance on what to extract or update

Workflow

1. Identify Target File

If filename provided:

• Check if file exists at given path (absolute or relative to project root)
• If not found, search common locations: project root, .claude/ directory

If no filename:

• Default to CLAUDE.md in project root
• If not found, check for agents.md or similar context files
• If no context file exists, offer to create CLAUDE.md

2. Analyze Conversation

Review the conversation history to identify:

User Information

• Personal preferences (coding style, tools, frameworks)
• Project-specific details (architecture decisions, conventions)
• Team/organization context
• Communication preferences

Technical Context

• Technology stack details
• File structure patterns
• Naming conventions
• Testing approaches
• Deployment configurations

Workflow Preferences

• How user likes to receive information
• Level of detail preferred
• Confirmation preferences
• Common tasks or patterns

3. Read Existing Context

Read the target file to understand:

• Current structure and sections
• Existing information (avoid duplicates)
• Writing style and format used

4. Determine Updates

Compare new information against existing context:

• Add: New sections or details not currently captured
• Update: Existing sections that need refinement or correction
• Skip: Information already present or not worth persisting

If user provided instructions, prioritize extracting that specific information.

5. Apply Updates

Preserve existing structure while integrating new information:

• Match the existing formatting style
• Add new sections at logical locations
• Update existing sections in-place when refining
• Use clear, concise language suitable for AI context

6. Confirm Changes

Show the user:

• Summary of what was added/updated
• The specific changes made (diff-style or list)

Guidelines

What to persist:

• Stable preferences and patterns (not one-time requests)
• Project architecture and conventions
• Recurring workflows or processes
• Corrections to previous assumptions

What NOT to persist:

• Temporary or session-specific details
• Sensitive information (credentials, personal data)
• Information already well-documented in code
• One-off task instructions

Writing style:

• Use imperative form for instructions
• Be concise—context files share limited token space
• Structure with clear headers for scannability
• Include examples only when they add clarity

Examples

Basic usage

User: /update-context

→ Scan conversation, identify learnings, update CLAUDE.md

Specific file

User: /update-context agents.md

→ Update agents.md with conversation learnings

With instructions

User: /update-context claude.md focus on the database schema we discussed

→ Extract database schema details and add to claude.md

After learning preferences

[Conversation where user explained they prefer TypeScript and functional patterns]
User: /update-context

→ Add coding preferences section to CLAUDE.md:

## Coding Preferences
- Use TypeScript for all new code
- Prefer functional patterns over classes
- Use named exports