Source Intelligence
Follow Discord

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

← All of v2.1.66 Home All releases ←olderv2.1.64 v2.1.67newer→
Claude Code v2.1.66

Sub-Agent Context Forking (forkContext)

What

Agents can now receive the full parent conversation context when spawned as sub-agents, enabling better awareness of what the user has been working on.

Details
  • New forkContext frontmatter field for agent .md definitions (set to true or false)
  • When enabled, the sub-agent receives prior conversation messages as read-only context
  • Agents with forkContext: true must use model: inherit to avoid context length mismatch
  • A new ### FORKING CONVERSATION CONTEXT ### boundary marker is injected to clearly separate inherited context from the sub-agent's own task
  • The sub-agent is instructed to only use tools available in its own system prompt, even if context messages reference other tools
Evidence

Sub-agent entry system (search for "sub_agent_entered" and "FORKING CONVERSATION CONTEXT")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

See this entry in the whole of v2.1.66 →