g-tasks

Behavior

1. Check prerequisites

Check if docs/features/ contains any feature directories. If not, stop: "No feature found. Please run /g-prd first."

If there are multiple features, list them and ask which one to generate tasks for. If there's only one, use it.

Check if the selected feature has a PLAN.md. If not, stop: "No plan found in docs/features/[name]/. Please run /g-plan first." If the feature has a PRD.md, read it for user story details and testing decisions. If it doesn't, warn: "No PRD found. Proceeding, but acceptance criteria may be less detailed." Read AGENTS.md if available. Read CONTEXT.md at the repo root if it exists — task descriptions should use the project's canonical vocabulary. CONTEXT.md is read-only at this stage: do not write to it, do not add new terms here (new terms surface in /g-prd or /g-plan interviews, not in tasks).

Verify-and-switch to the feature branch. Before any file write, ensure the working tree is on the feature branch {feature-name}. The switch is silent — invoking the stage implies the operator's intent.

Run this block literally. Do not paraphrase, shortcut on observation ("I can see we're already on the feature branch"), or skip when the current branch appears correct. The explicit run IS the spec compliance signal.

expected="{feature-name}"   # from selected feature directory
current=$(git rev-parse --abbrev-ref HEAD)

if [ "$current" = "$expected" ]; then
  : # already on it
elif git show-ref --verify --quiet "refs/heads/$expected"; then
  git checkout "$expected"
else
  echo "feature branch '$expected' missing — run /g-prd to create it"
  exit 1
fi

If the feature branch doesn't exist, STOP without modifying any file.

2. Parse the vertical slices

Read the Vertical Slices section from the feature's PLAN.md. Each slice becomes one task. Read the full plan for architectural context that needs to be embedded into each task. Read the PRD for user story details and testing decisions.