---

type: adr id: adr-0075 status: accepted created: 2026-06-13 updated: 2026-06-13

ADR-0075 — The starter kit ships as its own template repo; essential guides ship installed

Context

ADR-0069 kept the kit inside this repo "until public launch" so format edits stayed one-repo commits; ADR-0074 split the producer workspace (corpus-works) and the optional guides (corpus-skills) out, tiering all twelve advanced guides into the catalog. Two corrections from the owner followed. First, the kit split's precondition is gone: with the workspace in corpus-works cutting and reviewing every change, a format edit is already a reviewed cross-repo operation, and the family's repos exist. Second, ADR-0074's tier line was drawn in the wrong place: the workspace authoring guides (write-audit, write-research, write-rfc, write-prd, write-bug-report, write-change-plan, write-inventory, spec-check, split-work, save-findings, adversarial-review) author and gate corpus's own artifacts — every corpus workspace needs them, which is the definition of essential, not optional.

Decision

1. The kit is its own template repo: corpus-starter-kit. The repo root is the workspace (copy whole / "Use this template"); kit content is edited there. This repo's docs and ADRs remain the rule canon; the kit ships what the ADRs freeze. starter-kit/ leaves this repo; every doc reference points at the kit repo (URL form — the docs are read on GitHub).
2. Essential guides ship installed. The kit's .agents/skills/ carries the core loop (write-spec, implement-task, review-output) plus the eleven workspace authoring guides. advanced/ keeps the optional templates and the two reference cards.
3. The catalog is the optional layer. corpus-skills carries the six conditioning stances (persona-architect, -auditor, -documentarian, -researcher, -skeptic, -surveyor), the standalone empirical-proof, and the ten per-change-shape implementation guides — plus, under its docs/, the skill-design research layer (activation, body anatomy, execution, self-containment, task files, scope, sources) imported from the owner's upstream skills library.

Alternatives considered

• Keep the kit in-repo until launch (ADR-0069 as written) — its rationale (one-repo format edits) no longer holds: the workspace already makes every format change a planned, reviewed task, and a template repo gives adopters one-click copy instead of a subfolder extraction.
• A generated mirror (kit stays canonical here, repo derived) — doubles every edit and needs sync automation nobody has; the family precedent (ADR-0074: content edited in its shipping repo, rules canon here) already works.
• Essential guides as kit advanced/ opt-ins — "essential but uninstalled" is a contradiction adopters pay for; the DX audit's cold-adopter evidence says uninstalled guides simply don't get used.

Consequences

Accepted. Refines ADR-0064 (the optional tier is now: advanced templates + cards in the kit, stances + code depth in the catalog), ADR-0069 (the workspace shape stands; the defer-to-launch clause is executed; the kit-skill-surfacing note retires with the nested copy), ADR-0073 (the kit repo joins the governed family as derived content), and ADR-0074 (Decision 2's tier line moves: workspace authoring guides are kit content, not catalog content). Cost accepted: a format change is a two-repo change, mitigated by corpus-works review.

Propagation

AGENTS.md, README, docs/01/02/03/04/05/06/08/09/10, ADOPTING, examples ×3, reference/{artifact-formats, cheatsheet, memory, step-bars, review-stances, agent-guides, advanced-lifecycle}, checks/README, SKILLS-MANIFEST, ledger row; corpus-cli's scaffold resync; corpus-works pointer updates.