From the brief, to the build, to the receipts.

01 · THE BRIEF

The brief was a single 1,700-line markdown file.

AEGIS_Forge_Web_Brief_v1.0.md, prepared by Creative & Engineering Direction at OpenStateStack Solutions Europe PC. Every page, every CTA, every colour token, every comparison table — specified before any code was written.

02 · THE AGENT STACK

Claude Code Opus 4.7 supervised by one accountable human lead.

Claude Code (the CLI) running Anthropic's Claude Opus 4.7 (1M context window), with the Bash, Read, Edit, and Write tools. Filesystem-native; commits and decisions persisted as we worked.

03 · THE SUPERVISION

Triad supervision, one accountable lead.

Business Analyst, Product Manager, Senior Developer agent roles supervised the build at each phase boundary; the human lead reviewed every chapter boundary and signed off Sprint 0 before Sprint 1 began.

04 · WALL-CLOCK AND TOKEN SPEND

Recorded at QA, published here.

Total agent wall-clock, total tokens, total human review hours — all recorded in PROGRESS.md as we ship, and surfaced here at QA. Pre-publication clearance pending; see QUESTIONS_FOR_HUMAN.md item 7.

05 · WHAT WAS HARDEST

Restraint.

The hardest engineering decision was what NOT to do. No drop shadows. No gradients. No motion that does not earn its weight. The brand is institutional, not cinematic. Every commit deferred to that.

06 · WHAT WORKED

Treating the brief as binding spec.

Build order followed Section 15 of the brief; every commit referenced its section number; deviations were logged in DECISION_LOG.md, never silent.

07 · WHAT CHANGED OUR MIND

The deviations are listed in DECISION_LOG.md.

Three notable deviations: deploy target (Netlify replacing Vercel), repository layout (single Next.js app at root rather than pnpm workspaces), and content format (TypeScript modules in sprint 1, MDX migration in sprint 2). Each decision documented with the trade-off considered.