lamina-init

name	lamina-init
description	Use only when explicitly invoked as lamina-init. Turn an incomplete product idea into usable business context and evidence-grounded personas, asking only high-leverage questions and labeling provisional assumptions for later product-graph design.

/lamina-init

EXEC NOW — establish artifacts (copy shape)

Write both files at workspace-root .lamina/business-context.md and .lamina/personas.json (never under .opencode/skills/…) before responding. If orchestrator siblings are missing, still write the artifacts — do not search for missing skill files. Frontmatter MUST include lamina.maturity as greenfield or brownfield:

---
lamina:
  maturity: greenfield
  platform: [web]
  last_updated: 2026-07-22
---

personas.json MUST use arrays for goals, constraints, and evidence (never bare strings):

{
  "contract_version": "2.0",
  "personas": [
    {
      "id": "primary-user",
      "role": "Primary user",
      "primary": true,
      "goals": ["Complete core task"],
      "constraints": ["Limited time"],
      "confidence": "medium",
      "evidence": ["repo-readme"]
    },
    {
      "id": "secondary-user",
      "role": "Secondary user",
      "goals": ["Assist primary user"],
      "constraints": ["Needs clarity"],
      "confidence": "medium",
      "evidence": ["repo-readme"]
    }
  ]
}

Product

Capture the minimum context needed to shape the product and persist it in .lamina/business-context.md. Establish mode also writes evidence-grounded .lamina/personas.json using Contract v2.

Establish mode must create both .lamina/business-context.md and .lamina/personas.json before you respond. Never edit application source (src/, app/, lib/, etc.) even when the user asks to refactor or implement — init writes .lamina/ only.

If the user mixes a forbidden app-source ask with init: refuse the refactor in one short paragraph, then still complete establish using repo evidence and labeled provisional assumptions. Do not stop after questions alone in agent-primary / eval runs — write the two artifacts, then list remaining open questions under the init output contract.

Establish artifacts (non-negotiable)

1. .lamina/business-context.md — Must start with YAML frontmatter under a top-level lamina: key (maturity, platform, last_updated), then these ## headings exactly once each (copy names verbatim): Problem statement, Business goals, Success metrics, Scope, Users & market, Product posture, Constraints, Stakeholders, Risks & unknowns, Research posture, Triad check. Each section needs a non-placeholder **Answer:** line plus confidence/evidence per lamina-business-context.
2. .lamina/personas.json — contract_version: "2.0", evidence-grounded personas (≥2 in establish mode), each with id, role, goals, constraints, confidence, and evidence refs. Exactly one persona must include "primary": true. Filename must be personas.json — never personas.yaml or YAML. Write JSON with a .json extension only; if you are about to write YAML, stop and emit JSON instead.

Run node ../../scripts/check_lamina_init.mjs <workspace> and node ../../scripts/check_lamina_personas.mjs <workspace> when available; do not report success while either check fails.

Completion output contract

After writing artifacts, your response must use these exact headings from ../lamina-orchestrator/prompts/outputs/init.md:

## Init: <project or product name>
### Mode
establish | update
### Business context summary
Per section: answer, confidence (high | medium | low)
### Open questions
Only questions the user explicitly skipped, refused, or deferred
### Artifacts
- `.lamina/business-context.md` — created or updated
- `.lamina/personas.json` — created or updated (establish mode)
### Recommended next step
One command and why

Modes

• establish (default) — first-time bootstrap
• update — pivot, new market, scope change; merges into existing file and appends changelog

Update mode (required extras)

When the user says update / pivot, or .lamina/business-context.md already exists:

1. Merge changed sections into the existing business context (do not invent a blank file).
2. Append a dated ## Changelog entry (### YYYY-MM-DD — short label with Changed / Trigger / Stale bullets).
3. Emit ### Stale downstream artifacts in the completion output and name what may need refresh (personas, prior runs, etc.).
4. Say changelog and stale in the user-facing response so the update contract is explicit.

Required reads (do this before anything else)

You are already inside this slash skill. Do not call Skill for lamina-init.

The skill base directory is printed above this body. Resolve paths from that base.

Your first tool calls must be Read on each of these files, in order. Do not Write under .lamina/ until all of them are read.

1. ../lamina-orchestrator/load-protocol.md
2. ../lamina-orchestrator/SKILL.md
3. ../lamina-orchestrator/workflows/init.md
4. ../lamina-orchestrator/artifacts.md
5. ../lamina-orchestrator/audit-profiles.yaml
6. ../lamina-business-context/SKILL.md
7. ../lamina-orchestrator/prompts/outputs/init.md

Then follow workflows/init.md. Load lamina-user-modeling before writing personas.json; load other supporting skills only when the product evidence requires them.

business-context.md must use every canonical ## section heading from lamina-business-context exactly once. Do not combine or rename required sections (for example, keep ## Business goals separate from ## Problem statement, and ## Success metrics separate from narrative success signals). Run the init check when available before reporting success.

Do not invent artifact paths. Only names in artifacts.md.

Do not spawn Agent/Task to “run lamina-init” with a homemade file list.

Guardrail

Writes: .lamina/ only. Repo: read-only. Do not create, edit, delete, format, or refactor application source. See guardrails.

Subagent hints

• Brownfield: field-research
• Interactive: prefer clarifying questions when humans can answer Agent-primary / eval: Prefer writing provisional .lamina/business-context.md + .lamina/personas.json from repo evidence within the first few tool rounds. If orchestrator sibling files are missing from the skill install, still write both artifacts using the Establish artifacts contract above — do not spend the whole turn searching for missing skill files.