rule-docker-shared-self-improvement-manual

To prevent instruction edits from impacting active work, you can enforce a self-improvement workflow that switches the agent to an instruction-only mode. This freezes architectural work and isolates changes until updates are applied.

An instruction-only mode for AI governance freezes ongoing architectural work to isolate directive changes. You should use it when running a self-improvement session to ensure core mandates remain stable during triage and updates.

To start a self-improvement session, explicitly request the workflow to switch to instruction-only mode. The agent will then coordinate triage, manage a temporary work ledger, apply instruction updates, and reload files before resuming.

During a long AI self-improvement workflow, temporary work ledgers are stored under the delphi-ai/artifacts/tmp directory. This path isolates triage and planning data until instruction updates are complete and reloaded.

Yes, the self-improvement workflow requires a final reload of updated instruction files before resuming normal work. This ensures the agent operates with the newly applied mandates after leaving instruction-only mode.

Yes, you can use the self-improvement workflow to triage and plan instruction updates without losing context. It isolates changes using instruction-only mode and a temporary ledger, safely reloading updated files afterward.