Skip to content
Summon is temporarily deprecated. It is tuned for a version of Claude Code that no longer exists, and its multi-lens review does not fail loudly — it degrades into unanimous approval. Read why.

7-Phase Workflow

Summon’s methodology is phase-dependent: different agents lead at different stages, and the transitions between them are enforced.

PhaseLeadKey Pattern
DiscoveryCamElicit before implementing
ArchitectureArchieADR before code
PlanningPat + GraceWork items before sprints
ImplementationTara → SatoTDD red-green-refactor
Code ReviewVik + Tara + PierrotThree parallel lenses
Done GateGrace16-item checklist
Sprint BoundaryGraceRetro + velocity + kaizen

Cam leads. Before anyone writes code (including types, tests, or ADRs), three questions must be answered:

  1. Do work items exist? If not, create them.
  2. Is this an architectural decision? If yes, go to Architecture first.
  3. Am I about to write code? If yes, Tara writes tests first.

Archie leads, Wei challenges. Every feature with an architectural decision needs an ADR before implementation starts. The Architecture Gate requires:

  • Archie proposes a design
  • Wei pressure-tests it (devil’s advocate)
  • The ADR is written and approved
  • Only then does implementation begin

Pat writes acceptance criteria. Grace organizes into sprint waves: batches of work sized for one Claude Code session.

Strict TDD cycle:

  • Red: Tara writes a failing test
  • Green: Sato writes the minimum code to pass it
  • Refactor: Clean up while tests stay green

One commit per issue. Conventional commit format. Board status moves to “In Progress” before code starts.

Three parallel lenses, run every time:

  • Vik: Simplicity and maintainability
  • Tara: Test quality and coverage
  • Pierrot: Security and compliance

Critical or Important findings must be resolved before merge. Board status moves to “In Review.”

A 16-item checklist before any work item closes. Includes: tests pass, no security findings, docs updated, ADR written, board updated, and more.

Grace runs /sprint-boundary:

  • Retrospective: what worked, what didn’t
  • Velocity tracking
  • Kaizen: one concrete improvement for next sprint
  • Sweep: catch any missed items