Claude Code can now inject named context sections and a coordinator's worker-tool list into the system prompt
What
- Two new context-injection types,
coordinator_contextandcontext_sections, are added to the existing set of system-prompt injectors (environment, model, session_context, etc).coordinator_contextinjects the list of worker tools available to a coordinator in multi-agent ("coordinator/worker") setups, andcontext_sectionsinjects arbitrary named text sections (each with a name and text). - These same two injectors are also available as message attachments when building a conversation, with
coordinator_contextonly injected when a new gate check passes. - The session-refresh logic used when resuming a session now builds a flat list of
{name, text}sections (merged from user and system context) instead of separate fields, and always marks refreshed context asbare: trueon the error path instead of silently returning stale context. - A gate that used to suppress the environment/model/output_style/session_context/instructions/date/language injectors entirely has been removed, so these now always run.
Why This lets Claude Code pass structured, named context (like a coordinator's list of worker tools) into the system prompt in a more organized way, and makes session resumption more reliable by always refreshing context sections instead of occasionally reusing stale data.
tengu_indexed_corbato Not enough to sayNothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.268: off
Read once, for one account on one subscription tier, against v2.1.268. It isn't a statement about your account. What a flag value here can and cannot tell you
Whether this is switched on for any account is unknown; the `tengu_indexed_corbato` gate controlling it has not been read.
New in this build: tengu_indexed_corbato