Export Context

cortex export packages context into a form you can hand to another session, sub-agent, or external LLM. This is one of the most practical bridge workflows in Cortex.

What Export Is For

Use export when you need to:

  • hand context to a sub-agent
  • consult another LLM with a curated bundle
  • share only the parts of Cortex that matter for the current task
  • export specific agent definitions

Core Commands 

cortex export list
cortex export context -
cortex export agents code-reviewer

The main export categories exposed by the CLI are:

  • core
  • rules
  • modes
  • agents
  • mcp_docs
  • skills

Exporting Session Context

Write to stdout:

cortex export context -

Write to a file:

cortex export context review-context.md

Restrict the export:

cortex export context - --include core --include skills
cortex export context - --exclude mcp_docs --exclude skills

Exclude a specific file:

cortex export context - --exclude-file rules/quality-rules.md

Exporting Agents

Export one or more agent definitions directly:

cortex export agents code-reviewer
cortex export agents code-reviewer security-auditor --output review-agents.md

This is useful when the handoff is specifically about agent behavior rather than a broader session context bundle.

Agent-Generic vs Claude-Specific Output

Both context and agent export support:

--no-agent-generic

Use the default output when you want a general handoff format. Use --no-agent-generic when you specifically want the Claude-oriented format.

Export Workflow For Sub-Agents

One reliable pattern is:

# Build a compact context bundle
cortex export context - --include core --include rules --include skills > /tmp/task-context.md

# Optionally generate a skill shortlist too
cortex skills context --no-write > /tmp/skill-context.md

Then hand the exported files to the sub-agent or downstream workflow.

Export Workflow For External LLM Consultation

When consulting another model:

  1. export only the categories that matter
  2. remove or avoid sensitive data
  3. pair the export with a focused prompt
  4. treat the response as advisory, not authoritative

This is especially useful alongside the multi-llm-consult skill.