sd onboard — write a seeds section into the project’s CLAUDE.md / AGENTS.md, so every future agent session learns the tracker exists without anyone remembering to say so.
It is the persistent half of the pair whose ephemeral half is seeds-prime: prime renders the briefing per session on request, onboard commits it to the memory file the harness auto-loads. The choice of both files is what makes seeds harness-portable without naming harnesses — CLAUDE.md for claude-code, AGENTS.md for the cross-harness convention that opencode and others read.
The seeds repo dogfoods the result, and its own AGENTS.md shows what a fully-onboarded project looks like: a nine-step loop (prime → find unblocked work → claim → decompose if large → make focused changes → run the gates → pin debt markers → close & sync → record insights) with the sd commands inline at each step. That file is also the closest thing seeds has to a written methodology, which is part of why the framework classification holds at all.
Maps to no canonical SDLC stage — it configures the agent’s environment rather than advancing work. It produces artifact-standards in the narrow sense that the memory file it edits is where a project’s agent-facing conventions live.
See Also
- seeds-prime — the per-session counterpart.
- seeds-issue-workflow — the deeper, skill-shaped version of the same instructions.
- claude-code · opencode — the memory-file conventions it targets.
- artifact-standards — what it writes into.