A living project

Roadmap

This page makes the work visible: what was requested, what has evidence behind it, what is complete for this version, and what deserves a later refresh.

Milestones

StatusMilestoneEvidence and outcome
CompleteInspect the workspaceThe folder was empty apart from AGENTS.md; no application-specific conventions or toolchain were assumed.
CompleteLearn from prior workPast Codex work was reviewed for interaction patterns and reusable lessons: narrow skill routing, separate skill validation, boundary checks, browser acceptance, and honest baseline reporting.
CompleteResearch current guidanceCurrent OpenAI guidance on Codex instructions, skills, outcome-first prompting, testing, and agent orchestration was read; broader sources cover the open AGENTS.md format and trustworthy AI risk management.
CompleteReview review-skills firstReviewed the local meta-skills agent-harness-review, repository-audit, verification-gate, plus system review-agent and skill-creator.
CompleteBuild the handbookCreated interconnected static HTML pages, shared responsive styling, local copy buttons, accessible labels, and no external runtime dependency.
CompleteBrowser acceptance passOpened all 15 pages locally; verified titles, headings, stylesheet loading, no console errors, homepage → Prompting navigation, copy-button behavior, and no horizontal overflow at a 390px mobile viewport.
CompleteAdd exploration directionsAdded a linked starting map for media generation, local and hosted infrastructure, and multi-agent/autonomy experiments, with dated source anchors for volatile details.
NextPersonalize with real projectsAdd a small, dated project matrix for the repositories you actively use, keeping project-specific facts separate from general guidance.
NextRun a measured media baselineCompare one local image workflow with one rented or hosted path using the same prompt, asset, quality checks, time, and cost.
NextPrototype bounded delegationTry one read-only multi-agent review with explicit inputs, outputs, evidence, timeout, and human synthesis.
LaterMaintenance loopRecheck volatile model, Codex, skill, and plugin links when the product changes; record the date and reason for each refresh.
How to extend it

Keep the map small

Add a page when it has a distinct job. Put deep examples, project-specific commands, or a long reference list on that page instead of making the home page an encyclopedia.

How to close a milestone

Mark evidence, not intention

A milestone is complete when the artifact exists and the relevant behavior has been checked. “Planned,” “implemented,” and “verified” are different states.

What this roadmap does not promise

No guide can make every model reliable, replace human judgment, or stay current forever. Its job is to make good interaction patterns easier to choose and weak evidence harder to mistake for proof.