What problem does it solve? Agent-driven development loses context between sessions: settled decisions get re-litigated, in-flight work gets duplicated, and the user must verbally re-orient the agent every time. This skill performs a fast, durable read of project state so work can resume without a manual briefing. ## Core Features & Use Cases - Persistent-memory scan: Reads the ADR index, the active section of OBJECTIVES.md, and the Now horizon of ROADMAP.md to surface recent decisions, active objectives, and in-flight epics. - Live GitHub sweep: Runs batched gh queries for open PRs, blocker issues, needs-triage and needs-grooming counts, in-progress workstreams, and unlabeled issues, with a ~30 second performance budget. - Structured orientation summary: Prints a one-screen markdown report covering in-flight work, watchlist counts, recent ADRs, and 1-3 suggested next actions, without mutating any state. - Use Case: At the start of a work session, ask for orientation and receive a summary showing the in-flight epic, two open PRs awaiting review, five untriaged issues, and a suggested next action to confirm or override. ## Quick Start Ask the agent to orient you at the start of a session, for example by saying "session start" or "where are we?".
