seeds-issue-workflow

seeds-issue-workflow

Use this skill inside the seeds repo when you need to turn a vague ask into a tracked unit of work that sd ready will surface. It encodes two procedures: create-then-ready for a single scoped task, and plan-decomposition for work large or ambiguous enough that an LLM benefits from structured breakdown first.

Pre-flight

sd prime                          # inject rules + command reference
sd ready                          # see what is already unblocked
sd search "<keywords>"            # avoid filing a duplicate

If a seed already covers the work, claim it (sd update <id> --status in_progress) instead of filing a new one.

Procedure A — create-then-ready (single scoped task)

For a task that fits in one focused commit:

sd create --title "Fix sd close race on concurrent writers" \
  --type bug --priority 1 \
  --description "Two agents closing the same id can drop one update."

Capture the printed id (e.g. seeds-1a2b), then tag and wire it:

sd label add seeds-1a2b concurrency
sd dep add seeds-1a2b seeds-9f8e   # this work depends on seeds-9f8e first

A seed is ready when it is open and every id in its blockedBy list is closed. Add dependencies for real ordering only — every unnecessary sd dep add hides the work from sd ready. Verify:

sd ready | grep seeds-1a2b        # appears once its blockers close

Procedure B — plan-decomposition (large / ambiguous work)

When the work is multi-step, emit a structured prompt, fill it, and let seeds spawn the children:

sd plan prompt seeds-1a2b > /tmp/plan.json   # emit the template to fill
$EDITOR /tmp/plan.json                       # write one step per unit of work
sd plan submit seeds-1a2b --plan /tmp/plan.json

Each step becomes one child seed; a step's blocks: [j] wires step i's id into step j's blockedBy, so the children come out in dependency order. Steps may carry labels: [...] that flow to the spawned child (normalized and merged additively). Inspect and track: