Skip to content

/specd.continue

The primary command for driving a task through its full lifecycle. Reads your task’s current state, determines the next step, and executes it. One command from discussion through execution and review.

/specd.continue [task-name] [--interactive|--auto]

/specd.continue delegates to the brain — a config-driven orchestrator that reads pipeline.json and drives the lifecycle. It:

  1. Selects the task (from argument or shows a picker)
  2. Loads the pipeline configuration
  3. Reads current state and determines the next step
  4. Executes pre-hooks → step workflow → post-hooks
  5. Updates state and loops back

You never need to remember which stage comes next. The brain determines the next meaningful action from state.


ArgumentRequiredDescription
task-nameNoThe task to continue. If omitted, shows a picker for in-progress tasks.
--interactiveNoPrompt at every stage transition with skip/jump options
--autoNoRun everything without prompting, stop only on errors

ModeBehavior
Default (no flag)Auto-runs steps, pauses at natural boundaries (pause: true steps). Smart-skips steps that add no value.
--interactivePrompts at every stage transition. Shows skip and jump options. Recommends the natural next step.
--autoRuns the entire pipeline without prompting. Stops only on errors or task completion.

In default mode, the brain pauses before:

  • Execute — Lets you review the phase plan before implementation starts
  • Review — Review dispatches its own interactive flow

All other steps auto-run.

In interactive mode, you get explicit choices at each transition:

At discuss: Discuss (recommended) / Skip to research / Skip to planning At research: Research (recommended) / Skip to planning / Discuss more At plan: Plan (recommended) / Research first / Discuss more At execute: Execute (recommended) / Review plan / Stop for now

In default and auto modes, the brain skips steps that add no value:

  • discuss: skipped if no gray areas remain in CONTEXT.md
  • research: skipped for straightforward tasks (few files, clear requirements, no external dependencies)

discuss → research → plan → phase-execution loop
├─ phase-plan
├─ execute
├─ review
└─ revise (if needed)

Discussion — Probes open questions, records decisions. Context accumulates across sessions.

Research — Spawns 3 parallel agents: codebase integration patterns, external patterns, and pitfalls. Output: RESEARCH.md.

Planning — Creates ROADMAP.md with phases derived from dependency analysis. Phase goals only — no detailed PLAN.md files yet.

Phase execution — Nested sub-pipeline that loops per phase:

  1. Phase plan — Creates a detailed PLAN.md just-in-time from the roadmap goal
  2. Execute — Implements the plan, commits per task, tracks progress
  3. Review — Code review agent compares implementation against plan intent
  4. Revise — If review finds issues, creates fix plans (decimal phases like phase-01.1)

You can stop at any natural boundary. Progress is saved before each dispatch:

Progress saved. Pick up where you left off anytime:
/specd.continue user-dashboard

Resume from exactly where you stopped — the brain reads state and picks up at the right step.


When called without a task name:

/specd.continue

Specdacular scans .specd/tasks/*/config.json for in-progress tasks and shows a picker. If only one task exists, it’s selected automatically.


In a monorepo with orchestrator mode configured:

/specd.continue auth-system # Auto-picks next unblocked phase across projects
/specd.continue auth-system api # Target a specific sub-project

The brain respects cross-project dependencies from DEPENDENCIES.md and performs contract review after each phase to catch deviations before they cascade.


/specd.continue user-dashboard
/specd.continue payment-flow --interactive
/specd.continue api-rate-limiting --auto
/specd.continue