DivePlan

01 · Why DivePlan

In scuba, the plan is written before the diver goes under, then followed with discipline: plan your dive, dive your plan. That maps directly to this product. DivePlan turns intent into a staged, scored, auditable mission before execution, with the workflow modeled as briefing, descent, bottom time, ascent, safety stop, surface, and debrief. The name describes the operating discipline instead of falling back to another generic “forge” brand.

02 · Problem

My strongest AI builds were already multi-model: one system helped frame the product, another generated at high effort, and other coding agents hardened the result. The weakness was that I remained the human USB cable between every tab, paste, critique, score, retry, and handoff. The process could produce strong work, but it was fragile, difficult to audit, and easy to lose between sessions.

03 · Approach

1. 01

   Built a SwiftUI control plane for missions, prompt style, bounded loop cycles, scoring, skills, audit history, and packaging instead of restarting the product as another chat interface.

2. 02

   Kept the intended model roles explicit: Grok as strategist, critic, and optional live scorer; Google AI consumer sessions as the primary high-effort generator; official APIs as secondary fallbacks rather than quietly changing the product thesis.

3. 03

   Added assisted copy/open/paste delivery and a local Python companion with pair, risk-acceptance, and health gates. Automation fails closed when those gates are not satisfied.

4. 04

   Made quality measurable through a seven-dimension rubric, cycle caps, mission receipts, and a human approval gate before packaging or promotion.

5. 05

   Separated shipped code paths from owner proof: offline verification is automated, while device biometrics, real dogfood receipts, the final domain migration, and App Store release remain explicitly unclaimed gates.

04 · Outcome

In flight. Tier 1 Assisted AI Mode, the bounded loop/scoring path, local companion dry-run, audit, and packaging paths are implemented in the private repository. Device proof, dogfood missions, the move from the legacy loopforge preview URL to diveplan.io, and any App Store release are still open owner gates.