What problem does it solve? During an active Agentic SDLC run, accepted prompt revisions, spec changes, and new evidence accumulate without a consistent way to classify them or carry compact reminders into the next phase. This Skill refreshes the private STEERING.md ledger and machine-readable steering state so the coordinator can route the next phase with current context. ## Core Features & Use Cases - Steering Inbox Management: Appends each accepted prompt revision exactly once with a compact redacted summary, never storing raw prompt text, secrets, or customer data. - Disposition Classification: Classifies every unresolved entry as runtime-only, requirements-change, design-change, docs-update, resolved, superseded, rejected, or needs-human. - Routing Signals: Returns routing requests to sdlc-start (for example REPLAN_REQUIRED or HUMAN_INPUT_REQUIRED) instead of choosing the next phase directly. - Use Case: A user edits the managed prompt mid-run; the coordinator invokes this Skill to record the revision, classify its impact, and derive compact reminders before entering the implementation phase. ## Quick Start Ask the Agentic SDLC coordinator to refresh the active run's steering state from the latest accepted prompt revision before selecting the next phase.
