Crate roder_core

Re-exports§

artifacts
bus
conversation_forks: Conversation forks (roadmap phases 90 + 81).
dynamic_workflows
execution_lease: Local execution authority for a durable hosted-runtime owner. The supervisor extends this deadline only after a confirmed renewal of the same generation. Durable writes and remote/browser actions still require their own fencing.
external_tools: Metadata readback for host execution delivery recovery; no inputs or outputs.
fake_provider
forks: Runtime fork manager (roadmap phase 81, Task 2): resolves configured ForkProviders from the extension registry and applies path/policy checks before provider calls. Thread attachment lives in conversation_forks; these are the provider-facing primitives shared by conversation forks and the forks/* app-server surface.
hooks: Local Codex-compatible hook discovery and dispatch.
inference_routing
media_artifacts
media_generation: Provider-neutral image generation service.
model_profiles
policy_gate
review: Read-only code review sub-turns.
runtime
teams
tool_search: Runtime tool-search catalog adapter (roadmap phase 79, Task 2).

apply_agent_swarm_mode: Inject the agent-swarm reminder into the developer instructions while agent-swarm mode is active, so any app-server/SDK client (not just the TUI) nudges the model toward the agent_swarm fanout tool.
apply_codex_multi_agent_mode: Apply the effort-derived Codex V2 delegation policy. Ultra is a client-side orchestration mode layered over maximum model reasoning; lower efforts stay explicit-request-only so changing effort also changes delegation policy.
apply_model_instruction_overlay
apply_parallel_web_tools: Append Parallel search/extract guidance when those tools are on the turn.
apply_plan_mode
apply_runtime_profile
apply_task_ledger_required
apply_thread_developer_instructions: Prepends host-supplied thread instructions to the developer slot so they layer directly under the harness system prompt while harness addenda (runtime profile, plan mode, overlays) append after them.
apply_turn_developer_context: Sets the per-turn developer-context slot. Supplied on turn/start only and never persisted, so a context delivered on turn N is absent on turn N+1 unless the host sends it again. Providers render the slot after all stable instruction content — through a provider-native per-turn channel (e.g. a trailing system-role message) where available — so cached stable-prefix blocks survive per-turn changes.
default_instructions