sd prime [--compact] [--json] — emit the tracker’s rules and command reference into the agent’s context. --json returns typed sections so a harness can place them deliberately; --compact trims for a tighter budget.
It is step one of the framework’s prescribed agent workflow — “At session start run sd prime (and ml prime if mulch is the active context)” — and the reason it exists is that seeds is a program, not a skill: an agent cannot read sd’s instructions the way it reads a SKILL.md, so the tool has to inject them itself. That inverts the usual direction of pattern-context-engineering in this wiki, where the framework’s documents are loaded into the agent; here the framework hands over a rendered, versioned briefing on request, and --compact makes the cost of it a caller’s decision.
Its pattern-session-handoff edge is the pairing with seeds-sync: state committed to git on one side, primed back into a fresh agent on the other. Where gstack-context-restore and mp-handoff reconstruct a conversation, sd prime restores the queue — a fresh agent needs no narrative, only the rules and what is ready.
Maps to no canonical SDLC stage; it is session bootstrap, filed alongside the wiki’s other context-continuity capabilities.
See Also
- seeds-onboard — the persistent counterpart: written into the memory file once, rather than emitted each session.
- seeds-ready — the query a primed agent runs next.
- gstack-context-restore · mp-handoff · sp-using-superpowers — session-bootstrap counterparts elsewhere.