Skip to content

Changelog

All notable changes to this guide are documented here. Format loosely follows Keep a Changelog. The repo-root mirror lives at /CHANGELOG.md.

[0.6.1] — 2026-05-06 — Standalone Claude site

Changed

  • Claude guidance now lives at the separate Claude site instead of inside the Field Guide navigation.
  • Homepage, Mastery hub, examples, source audit, and Claude redirect page now point to the standalone site.
  • Removed Field Guide loading of the Claude-only CSS and JavaScript assets.

Verified

  • Claude source links were refreshed against official Anthropic docs, including the current canonical plan chooser URL.

[0.6.0] — 2026-05-06 — Claude Mastery microsite

Added

  • Claude Mastery — a beginner-first, step-by-step Claude setup and mastery microsite with a path picker, local progress checklist, copyable prompt kit, Projects walkthrough, feature map, 30-day practice plan, and safety baseline.
  • Scoped claude-mastery.css and claude-mastery.js assets. The checklist persists only in the visitor's browser via localStorage; copy buttons use the browser Clipboard API and make no network calls.

Changed

  • Homepage now surfaces Claude Mastery as a first-class starting point for readers learning Claude specifically.
  • The older mastery/claude.md track now points readers to the dedicated Claude Mastery microsite to avoid duplicated, drifting Claude guidance.

Verified

  • Claude feature claims rechecked against current official Anthropic sources for plans, usage limits, personalization, Projects, memory, incognito chats, Artifacts, Skills, connectors, Claude Code, and prompting best practices.

[0.5.2] — 2026-05-06 — Audit remediation and source refresh

Changed

  • Homepage now has a "Pick the right starting point" routing table for absolute beginners, subscription maximizers, product choosers, agent builders, and team/sensitive-data owners.
  • GitHub Copilot references now use the current Copilot cloud agent name, with "formerly Copilot coding agent" retained only where it clarifies vendor renaming.
  • Copilot cloud agent availability and billing guidance now reflects the current GitHub Docs: Pro, Pro+, Business, and Enterprise availability; Business/Enterprise policy enablement; premium-request usage per cloud-agent session and steering comment.
  • xAI model links now use the current canonical docs.x.ai/developers/models URL, and Grok API examples default to the currently documented grok-4.3 model identifier while still reading from XAI_MODEL.
  • The MkDocs Material header repo widget was removed to avoid unauthenticated GitHub API calls and possible 403 console errors in visitor browsers.
  • OpenAI recipe snippets now use OPENAI_MODEL instead of stale hard-coded model IDs, and beginner setup steps include the matching env var.
  • Exact live-cost claims were removed from recipes; cost sections now describe token drivers and tell readers to recalculate against current vendor pricing.
  • Structured-output language now tells readers to validate returned objects locally instead of treating provider-side schema constraints as a substitute for production validation.
  • Browser/computer-use comparison now marks computer-use-preview as legacy and reflects current OpenAI computer tool guidance.

Verified

  • Source refresh checked current official GitHub Copilot cloud-agent docs, GitHub Copilot requests/billing docs, xAI structured-output docs, xAI model docs, and OpenAI computer-use docs.
  • Site QA now guards against stale xAI model URLs, stale Copilot cloud-agent billing language, hard-coded older OpenAI model IDs, and exact live-cost claims.

[0.5.0] — 2026-05-06 — Subscription-first expansion (deepening pass)

Added

  • Capability map — cross-product matrix covering 18 capability rows across Claude / ChatGPT / Gemini / Grok / Perplexity / Copilot / "any AI tool" with plan-availability tags and per-cell fallbacks.
  • No-code automation guide — ChatGPT Tasks, Gemini scheduled actions, Claude Cowork, calendar-driven manual fallback, plus the safety drill (3 manual runs, off-switch, cost cap).
  • Memory & preferences guide — portable AI profile, global vs project memory, migration playbook, 6 copy-paste templates.
  • Examples library — 13 beginner-to-expert guided examples, each with a 6-layer ladder (chat → project → memory → native task → custom assistant → dev/API).
  • Universal 7-mission learning path — the no-API on-ramp from "I have a subscription" to "I'm getting full value."
  • Plain-English glossary with schematic SVG diagrams for ChatGPT Project & Task, Claude Project / Skill / Cowork, Gemini Gem / Scheduled action, Grok chat.
  • Task Builder v0.5: 4-tier mode selector (Beginner / Power User / Builder / Developer), 7 new output panels (Use this in…, Learning ladder, What good output looks like, Practice exercise, Free vs paid, Maximize my subscription, Expert expansion). Provider packages render for all 7 surfaces with the user's primary app first.
  • Repo-root SUBSCRIPTION_FIRST_EXPANSION_STATUS.md and USER_REQUIREMENTS_TRACEABILITY.md extended; docs/product-requirements/user-requirements-traceability.md mirrors them.

Changed

  • Homepage card grid expanded with Learning path, Capability map, No-code automations, Memory, Examples.
  • mkdocs.yml nav re-ordered: Task Builder → Learning path → Mastery → Capability map → Examples → No-code automations → Memory → Surface router → (rest unchanged).
  • Roadmap rewritten to reflect the subscription-first strategy and what landed in v0.5.
  • Source map gains a v0.5 "Subscription-first vendor surfaces" section; source audit gains a v0.5 audit entry.

Verified

  • mkdocs build --strict exits 0.
  • Beginner / Power User / Builder modes never recommend API / CLI / MCP / local scripts.
  • Task Builder JS contains no fetch, XMLHttpRequest, WebSocket, sendBeacon, dynamic script loading, or analytics calls (only commented references to those names exist as guardrails).
  • All runnable scripts in starter kits remain gated behind RIZWAN_APPROVED_TO_RUN=1.
  • All 16 user requirements and all 24 acceptance criteria are mapped to implementations in the traceability files.

[0.4.0] — 2026-05-06 — Subscription-first expansion

Added

  • Mastery section: index plus 7 beginner-to-expert tracks for Claude, ChatGPT, Gemini, Grok, Perplexity, Coding agents, and "Any AI tool / I don't know."
  • Task Builder Beginner Mode (the new default) plus a primary-app picker, a comfort-level picker, a 16-option job picker, and two new output panels: Level up this workflow and If your plan doesn't have this.
  • A What to click output panel with exact step-by-step instructions per AI app.
  • User requirements traceability page mapping every user requirement to its implementation.

Changed

  • Homepage opens with "No API key required. Start with the AI app you already use."
  • Task Builder page rewritten with primary-app + comfort + job at the top.
  • Advanced surface picker is hidden in Beginner Mode.

Verified

  • mkdocs build --strict exits 0.
  • Beginner Mode never recommends API/CLI/MCP/local scripts.
  • Task Builder JS contains no network calls.

[0.3.0] — 2026-05-06 — Task Builder transformation

Added

  • Task Builder — a static, browser-local tool that turns a task description into a recommended surface plus a complete package (prompt, system instructions, memory block, setup, tool allowlist, HITL gates, evals, red-team probes, repeat-run playbook, troubleshooting, exportable JSON).
  • Surface router — long-form decision matrix for chat vs. project vs. coding agent vs. skill vs. automation vs. agent vs. browser-use vs. MCP.
  • Grok / xAI coverage — platform page, quickstart, two recipes (structured outputs and tool calling). Distinguishes consumer Grok, Grok on X, and the xAI API.
  • Model freshness page — env-var pattern (OPENAI_MODEL, ANTHROPIC_MODEL, GEMINI_MODEL, XAI_MODEL) and a quarterly drift checklist.
  • ADR 0004 documenting the recipe nav pattern.
  • IMPROVEMENT_PASS_STATUS.md at the repo root.

Changed

  • Homepage rewritten as a task-to-outcome on-ramp.
  • Platforms index gains an xAI / Grok row.
  • Source map and source audit include xAI / Grok.

Fixed

  • MCP starter-kit approval bypass removed from both Python and TypeScript MCP servers; RIZWAN_APPROVED_TO_RUN=1 is now unconditionally required.
  • Deployment claims corrected in DEPLOYMENT_STATUS.md, the root CHANGELOG.md, and this page.
  • Link check gating split into a gating internal-links job and an advisory external-links job.
  • Model currentness fixed in starter-kits/local-script-agent/script.py (now reads OPENAI_MODEL).

Verified

  • mkdocs build --strict exits 0.
  • The Task Builder JavaScript contains no network calls.
  • All v0.2 work preserved.

[0.2.0] — 2026-05-06 — Productionization pass

Added

  • Quickstarts — five end-to-end recipes (read-only, file-editing, coding, browser, MCP).
  • Recipe library — 40+ complete agent recipes covering common workflows.
  • Starter kits — nine copyable kits with specs, prompts, allowlists, eval/red-team JSONL, and launch checklists. Example scripts are inert by default.
  • Eval banks — 100+ golden cases and 100+ red-team cases with rubrics and per-platform usage notes.
  • Agent Factory module — a 25-candidate → 5-shipped worked example.
  • Source audit — a record of which pages were rechecked against which official URLs.
  • Repo hygiene — issue/PR templates, Dependabot, PR build workflow, link-check workflow.
  • Deployment docs — DEPLOYMENT_STATUS.md and PRODUCTION_DEPLOYMENT.md at repo root.

Changed

  • Removed all "(coming soon)" labels from the Platforms index — every page exists and is linked.
  • Nav label Safety/safety-checklists corrected to Safety checklists.
  • Source map expanded with additional Anthropic, OpenAI, Google, GitHub, and MCP references.

Verified

  • mkdocs build --strict exits 0. The build emits an upstream Material for MkDocs informational banner about the future MkDocs 2.0 release (theme-level, not a project warning) and INFO-level notices for the 41 recipes accessed through the Recipes index — see ADR 0004.
  • GitHub Pages production deployment succeeds.

[0.1.0] — 2026-05-06

Added

  • Initial public release of the AI Agent Builder Field Guide.
  • Start Here section: what an agent is, good/bad tasks, decision tree, four setup paths, safety baseline.
  • Platform setup guides for Claude Desktop, Claude Code, Claude Projects, Gemini app, Gemini CLI, Antigravity, Google AI Studio, ChatGPT, Custom GPTs, OpenAI API/Agents SDK, Codex CLI, GitHub Copilot cloud agent, and local scripts/schedulers.
  • MCP section: concepts, installation, writing servers, remote/custom connectors, security.
  • Browser & computer use section: Anthropic, OpenAI, browser-use, operating boundaries.
  • Orchestration section: single-agent loops, multi-agent patterns, local-first, state/memory.
  • Evaluation & safety section: eval sets, red-team workflows, safety checklists, HITL, incident response.
  • Templates: agent spec, prompt, eval rubric, safety checklist, agent PRD.
  • Reference: glossary, source map, roadmap, decision records.
  • MkDocs Material build, GitHub Actions deploy workflow, MIT license, CONTRIBUTING, CODE_OF_CONDUCT, SECURITY.

Source verification

  • All product pages verified against the official sources listed in source-map.md on 2026-05-06.