Skip to content

0001 — Shape Up over Scrum for solo + AI development

Architecture Decision Record.

Status

Accepted

Date: 2026-05-28

Context

This scaffold’s primary developer is one person, sometimes working with Claude Code sub-agents (prometeo, forja, centinela). The Inceptor workflow already in place — issue → PR → merge, orchestrated by the main Claude Code session — needs a cadence framework on top of it.

Scrum requires team ceremonies (standups, sprint reviews, retros with peers) that don’t function with N=1. Shape Up (Basecamp, 2019) was designed for small teams running fixed-time-variable-scope cycles. Its primitives translate cleanly to solo + AI work.

Alternatives considered:

  • Scrum-lite (no standups, but keep sprints + retros) — still ceremony for one person; sprints encourage fixed-scope which fights well with AI-assisted velocity variance
  • Kanban — no cadence at all; loses the “appetite” discipline that prevents scope creep
  • Ad-hoc — what we’d been doing; degraded into the 17-item deferred backlog the integration shipped with

Decision

Use Shape Up’s cadence primitives:

Shape UpThis repo
PitchA written issue spec — INTEGRATION-PLAN.md entries or .github/ISSUE_TEMPLATE/ forms
AppetitePhase or milestone (fixed time, variable scope)
Betting tableMonday ritual — pick the next pitch, hand the scope to Claude Code
CycleThe phase duration
Cooldown2-day gap after each phase for docs, refactors, dep bumps, debt

Agile Manifesto values stay as taste guiding choices, not ceremonies on a calendar.

Standups, sprint reviews, sprint retros are dropped. Lessons-learned go to docs/lessons.md only when centinela rejects the same issue twice in a row — specific trigger, low overhead.

Full framework: docs/PRINCIPLES.md §1.

Consequences

Positive:

  • Zero ceremony overhead for solo work
  • Pitches force explicit appetite + boundaries up front (better than open-ended sprints)
  • Cooldown gives explicit space for non-feature work (the 17 deferred items from prior phases live there)
  • Cleanly maps onto the sub-agent roles already in place

Negative:

  • “Betting table” with one person is a self-discipline ritual, easily skipped
  • No retro cadence means lessons can be forgotten — mitigated by the rejection-trigger rule, but not perfectly

Neutral:

  • Sub-agents (prometeo / forja / centinela) absorb roles that would otherwise be PO / dev / QA — Shape Up doesn’t prescribe these roles, so the mapping is clean and unforced

Supersedes

None — first methodology ADR.

References

Report an issue

Show diagnostics

Environment

 

Console errors

 

Failed requests