Add `--output json` flag for `init` and `add` commands

[lloydrichards]

Parent

• JSON/Structured Output Mode for Plan #102 (JSON/Structured Output Mode for Plan)

What to build

Add an --output json flag to both init and add commands. When provided (alongside --yes), the command runs the full pipeline silently (no interactive prompts, no terminal rendering), writes files to disk, and returns a structured ApplyResult as JSON on stdout.

This is the deterministic execution path for LLMs — used for greenfield scaffolding where conflicts aren't expected and the LLM just wants the CLI to do its thing and report back.

Output Shape (example)

{
  "ok": true,
  "created": ["src/db/index.ts", "src/server/main.ts"],
  "modified": [],
  "skipped": [],
  "failed": [],
  "finalize": [
    { "label": "Install dependencies", "command": "bun install" }
  ]
}

LLM Workflow Context

schema (#109) → construct Selection → plan (#102) → apply with own judgment
                                    OR
schema (#109) → construct Selection → add --yes --output json (this) → done

Acceptance criteria

• --output json flag accepted on both init and add
• When combined with --yes, runs full pipeline with no interactive prompts
• All terminal rendering suppressed (no Ink output)
• Structured ApplyResult JSON printed to stdout
• Finalize scripts included in output (not executed unless --trust also passed)
• Exit code reflects success/failure

Blocked by

• JSON/Structured Output Mode for Plan #102 (Plan command — shares serialization infrastructure)