# Build Week Submission: Sidequest ## One line **Sidequest turns a fresh Codex task into a guided apprenticeship—and makes the app teaching you become your first real Codex project.** ## The problem Codex can inspect folders, build apps, work with files, run long goals, delegate to agents, use connected tools, and support everyday work. A new user still faces an empty composer and a model picker. They do not know which folder to open, which model or effort to choose, how tokens work, whether a follow-up will steer or queue, when to open another task, or how to verify an edit. Documentation explains features. It rarely gives a beginner a safe sequence of real actions that compounds into independent practice. ## The solution Sidequest is a forkable repository plus a live companion app. The repository teaches Codex how to teach: 1. The learner opens the folder in a fresh Codex Local task and says “Start Sidequest.” 2. `AGENTS.md` invokes a project-local instructor skill. 3. Codex runs a doctor, launches a zero-dependency live-reloading app, and teaches one mission at a time. 4. The first missions cover the actual Codex cockpit: sidebar app, environment, models, tokens, steering, side chats, queues, plans, goals, and Git boundaries. 5. Codex maps the repo without editing. 6. The learner asks for an 8-bit style or new language. Codex edits the real learner source; the app beside it reloads visibly. 7. Deterministic checks, browser evidence, and explicit learner proof unlock the next mission. 8. The learner progresses through visible new tasks, task handoffs, pets, subagents, everyday files, reusable skills, connected tools, Sites planning, safe automation design, and a personal capstone. 9. Graduation happens in a fresh task. If the workflow still works without the tutorial conversation, the learner keeps the capability. The sequence is mastery-gated rather than completion-themed. Every mission exposes observable learning outcomes, a practicum, an aha moment, and an advance gate. The learner records practice and reflection in the app; Codex runs the workspace verifier after the learner supplies evidence, and only then does the next mission unlock. The Atlas view turns that curriculum into a navigable dependency graph. Locked, available, practice-logged, and gate-passed states are visible; selecting a node reveals its prerequisites and practicum. It is a map of what to learn next, not a passive documentation wall. ## The wow moment The learner selects “8-bit arcade” in Sidequest. Nothing visual happens. They copy the mission into Codex. Codex explains the scope, edits `learner/theme.css`, and the open app instantly reloads into the learner’s version. The same app continues to teach, now visibly changed by its student. A smoke test and browser pass earn the stamp. This is not a prebuilt theme toggle. The source edit, dev-server reload, and proof are the lesson. Mission 07 is a lesson, not a pet download. The bundled Cody sprite is a fallback fixture for people without Codex. Codex learners choose the desktop App-only pet path or ask Codex to “make me a friend,” then use imagegen for a personality concept, bounded subagents for grounded sprite rows, `hatch-pet` for a validated 8×9 atlas, and a real code edit that replaces Cody with their own helper and makes it react to mission and proof state in the open app. ## The Build Week extension The pre-extension project is preserved at commit `62af3ef`, tagged `pre-5.6-extension`. Build Week added Missions 21–26 as one evidence-forward Ship Loop: 1. **Prompt Clinic** turns a vague request into an outcome, limits, and evidence. 2. **Ship with Sites** separates compatibility planning, reviewable version state, deployment, access, and hosted verification. 3. **Browser Proof** verifies the served identity, critical journey, persistence, responsive layout, keyboard behavior, accessibility, and console health. 4. **Safe Automations** classifies risk and runs a deterministic local simulation without activating recurring work. 5. **Package the Workflow** turns the proven sequence into the project-local `$sidequest-ship-loop` skill. 6. **Final Flight** gathers provenance into a local practice credential with explicit limitations. Judge Mode presents that loop with isolated, clearly labeled sample evidence. It cannot modify learner progress, create an automation, or convert local practice into hosted proof. ## Why Codex is essential Codex is not merely how the repository was authored. It is the runtime instructor and the medium of learning: - repo-scoped behavior from `AGENTS.md` - a reusable project-local coaching skill - a challenge contract for fixtures, validators, hints, resets, and fresh-task tests - live local file inspection and edits - terminal processes and deterministic checks - browser verification - steering, queues, plans, and long-running goals - visible tasks and context handoffs - subagent delegation and synthesis - durable transfer into `AGENTS.md`, a skill, or another workflow surface The app deliberately avoids embedding another API chat. Learners practice the native product they are trying to learn, and the core path needs no separate API key. ## Technical implementation - Zero-dependency HTML, CSS, and JavaScript app optimized for a narrow side-by-side viewport - Node standard-library development server with a loopback-only Host boundary, an explicit public-file allowlist, range requests, traversal protection, SSE live reload, health endpoint, and read-only progress endpoint - Local proof/state CLI with deterministic mission checks and explicit learner-evidence gating - Backward-designed mission gate spine with observable outcomes, real practica, retrieval reflections, visible aha moments, and Codex-verified progression - `localStorage` for immediate browser continuity plus `.sidequest/state.json` for Codex-written verified progress - Project-local skill with a coaching state machine, hint ladder, authority ladder, and mission-specific invariants - Safe learner source seams and fictional file fixtures - Generated Cody atlas, manifest, contact sheet, and validation receipts wired to live app state - Four adaptive routes: Build, Work, Learn, and Life - Keyboard command menu, prompt lab, accessibility focus states, reduced motion, and narrow sidebar layout - Deterministic Ship Loop modules for prompt scoring, Sites readiness, browser verification, automation safety, workflow packaging, Final Flight, and isolated Judge Mode state - Cloudflare-compatible Sites build with a verified public deployment at `https://sidequest-ship-loop.qzhum1996.chatgpt.site` ## Judging fit ### Technological implementation Sidequest uses Codex’s native workspace, instruction, skill, task, goal, subagent, terminal, and browser patterns as one coherent product. Progress is backed by live source changes, deterministic validation, state events, and learner-confirmed proof rather than decorative course completion. ### Design The app is a focused field manual, not a dashboard or fake chat. One mission occupies the canvas; proof sits beside it. It works as a full page and in the narrow companion width where the concept matters most. ### Potential impact The audience is every first-time Codex user who knows the product is powerful but does not know what to ask first. The first useful loop takes roughly 30 minutes and teaches transferable product behavior before advanced setup. Schools, teams, communities, and creators can fork the mission pack for their own onboarding. ### Quality of the idea The course and its subject are the same artifact. Codex teaches the learner by inspecting, changing, and eventually replacing the tutorial with the learner’s own durable workflow. The visual live remix makes agency obvious; the fresh-task final exam proves transfer. ## 2½-minute demo arc 1. Establish the beginner problem and the original local apprenticeship. 2. Introduce Sidequest 5.6 and the six-stage Ship Loop. 3. Show Prompt Clinic’s before/after evidence and the Flight Recorder provenance trail. 4. Show the real Sites **Practice Complete** receipt without implying a deployment. 5. Show Browser Proof’s verified-local receipt, including the preserved identity-version defect and successful retest. 6. Show Safe Automations rejecting the destructive scenario with explained blockers. 7. Show the validated local packaging receipt for the `$sidequest-ship-loop` skill. 8. Show Final Flight’s clearly labeled sample practice credential. 9. Close on the readable desktop Judge Mode opening and invite a fresh-task replay. Closing line: **“This is not a course about Codex. It is a course made out of Codex—and it ends when you no longer need the course.”** ## Submission checklist - Working/runnable project: complete locally; use `npm start` and verify `/__sidequest/health` - Public code repository: `https://github.com/Fishbiscuit/sidequest`; `build-week-5.6` and `build-week-submission` are synced at the final documentation commit - Hosted app: `https://sidequest-ship-loop.qzhum1996.chatgpt.site` - Demo video: public YouTube link https://youtu.be/Bt6icgzZabI; local 2:30 MP4, narration, captions, transcript, and shot list remain in `submission/video/` - Project description: final copy in `submission/devpost-copy.md` - Screenshots: final set in `submission/screenshots/`; runthrough evidence in `assets/runthrough/` - Required model evidence: entrant must provide the primary Codex `/feedback` Session ID that verifies GPT-5.6 use - Test command: `npm test` - Final action: entrant reviews the form and submits before the deadline ## Challenge notes The live rules were rechecked on July 21, 2026. The published deadline is July 21, 2026 at 5:00 PM PDT. The demo must be under three minutes, include audio, explain what was built and how Codex/GPT-5.6 were used, and be publicly visible on YouTube. The submission also needs a functioning project link, repository access and license, an honest existing-project disclosure, and the primary `/feedback` Codex Session ID. Sources: [OpenAI Build Week](https://openai.com/build-week/), [Devpost challenge and judging criteria](https://openai.devpost.com/).