Not every change needs a multi-phase project. A bug fix, a rename, a small feature toggle — these are well-scoped enough that the overhead ofDocumentation Index
Fetch the complete documentation index at: https://docs.joggr.ai/llms.txt
Use this file to discover all available pages before exploring further.
/gg-new with separate discuss, research, plan, execute, and verify invocations is unnecessary. /gg-task runs the entire pipeline in one session and produces two artifacts instead of a full project directory.
When to Use /gg-task vs /gg-new
/gg-task
Small, well-scoped changes. A bug fix, a rename, adding a config option. Single session, two
artifacts (
spec.md and plan.md), no separate phase files./gg-new
Multi-file features, cross-cutting refactors, work that spans multiple apps or packages. Full
project scaffolding with separate phases, review gates between stages, and team coordination.
/gg-task when:
- The change is small and well-scoped (a bug fix, a rename, adding a toggle)
- You do not need multi-phase planning or team coordination
- You want to go from idea to implementation without leaving the terminal
/gg-new when:
- The change spans multiple apps or packages
- You need multi-phase planning with separate review gates
- Multiple developers will collaborate on the implementation
The Six-Step Workflow
/gg-task moves through six steps in order. Each step updates the step field in spec.md frontmatter, so the skill always knows where it left off.
New
Task scaffolding. The skill derives a slug from your description, creates
.joggr/.gg/tasks/ {slug}/, and writes an initial spec.md with YAML frontmatter tracking workflow state.Discuss
Structured interview to clarify goals, constraints, edge cases, and acceptance criteria. The
skill asks targeted questions and iterates if answers are vague — a shallow discussion produces
a weak plan. Answers are appended to the Discussion section of
spec.md.Research
Codebase analysis. The researcher agent identifies relevant files, existing patterns, and
dependencies, then surfaces 2-3 implementation approaches. You pick the approach before anything
is committed to the spec. After research, the review UI opens so you can edit the spec before
planning.
Plan
Flat task list generation. The planner agent reads the approved spec and produces
plan.md with
a set of concrete tasks. The review UI opens again so you can approve or edit the plan before
execution starts.Execute
Parallel code implementation. Tasks are grouped into conflict-free waves and multiple executor
agents run in parallel within each wave. Validation (typecheck, lint, tests) runs after each
task. You checkpoint and confirm after each wave before continuing.
Starting a Task
Pass a description (rough idea is fine):add-retry-logic), creates the task directory, and walks through all six steps. You can also pass just a slug if you prefer:
Artifact Structure
After a full run, the task directory contains two files:spec.md is the evolving specification. YAML frontmatter tracks the current workflow step and status. The body accumulates discussion notes, research findings, and acceptance criteria as the workflow progresses.
plan.md is the task list. It always uses a <phases> wrapper — one or more phases each containing a task list — in a single file, never separate phase directories.
Resume Behavior
If your session is cleared or times out mid-workflow, re-invoke with the same slug:spec.md frontmatter and picks up where you left off. Completed steps are skipped. Steps interrupted mid-run are retried from the beginning, except execution — which resumes from the last incomplete task.
If only one task exists in .joggr/.gg/tasks/, the skill auto-resolves it — no slug needed: