claude-sdk-rs: Command for Claude Code

Prime — Deep orient to the current project at the start of a session.

For routine session starts, use /session-recap instead — lighter, and covers the common case. Reach for /prime for genuine deep re-orientation: first session in a repo, returning after a long gap, or when /session-recap's summary isn't enough to get moving.

Instructions

1. Read each file listed in Context / Files to Read in order.
2. Run git ls-files to see the full tracked file tree.
3. Check for an active handoff: run ls planning/handoff.md 2>/dev/null.
   • If the file exists, read it. Lead your summary with an Active Handoff section (see format below) before everything else. This is the most important thing a new session agent needs to see.
   • If absent, skip this section silently. 3.5. Freshness gate — mev validate-brain --sync. Check whether this repo participates in a brain (walk up from the current directory looking for brain.toml, same resolution /log-work uses). Degrade gracefully rather than failing:
   • If mev is not found on PATH, say so in the Freshness line and continue — do not treat this as an error.
   • If no brain.toml is found (standalone repo), say so in the Freshness line and continue.
   • Otherwise run mev validate-brain --sync (read-only) from the brain root. If it reports every watermark in sync, note that. If it reports a drift (a project's source timestamp does not match its cache's synced_from, e.g. E_SYNC_DRIFT / E_SYNC_WATERMARK_MISSING), name the stale project(s) and offer to run mev emit-state --write to reconcile them — never run it without the user's explicit confirmation (D26: validation stays read-only by default). If the user declines or doesn't respond, proceed with the rest of /prime unchanged; the drift is informational, not a blocker. 3.6. Warm memory — by budget, never by reflex (brain decision D67). Run wc -w planning/knowledge.md planning/memory.md and branch on the combined count:
   • ≤ 2,500 words → read both files in full.
   • over 2,500 → read only the topic list (grep '^### ' planning/knowledge.md planning/memory.md), then open a named section on request and only then.