start-task

To enforce task classification and vault loading before generating non-code artifacts, initialize a front-loaded discipline checklist that queries Tier-3 substrates and loads relevant vault subgraphs prior to any output. This prevents ghost writing by ensuring required context is never bypassed.

The best way to structure an artifact handoff plan for research syntheses and ADR drafts is to use a first-reply contract with explicit headings and sibling-skill deferral. This delegates actual production to the correct specialized sibling skill rather than outputting it directly.

You need council routing for external publications and higher-risk decisions when your non-code workflow warrants an adversarial second-opinion review. This enterprise guardrail ensures disciplined artifact generation by triggering targeted critical analysis before final handoff.

Vault subgraph loading for non-code workflows like decks, reports, and diagrams works by loading only the relevant docs/01_Vault/<ProjectKey>/ nodes based on the specific task at hand. This targeted loading ensures substrate defaults are properly applied rather than skipped.

Your model skips substrate steps and ghost writes missing context during artifact reviews because it bypasses vault loading without a front-loaded discipline checklist. Requiring an auditable stamp file and explicit task classification forces the model to query Tier-3 substrates before producing output.