Canonical page: https://website-build-skill.aunysillyme.dev/docs/overview/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/README.md Relative links below resolve against that repository source location. --- # website-build-skill [![npm version](https://img.shields.io/npm/v/website-build-skill)](https://www.npmjs.com/package/website-build-skill) [![Repository checks](https://github.com/aunysillyme/website-build-skill/actions/workflows/check.yml/badge.svg)](https://github.com/aunysillyme/website-build-skill/actions/workflows/check.yml) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE) **A skill pack that turns your AI into an expert website builder.** It teaches your LLM or agent current, researched practice in design, brand, accessibility, performance, search, security, stacks and code, so the sites it builds look and work like a professional made them. - **The problem:** an AI builds websites from what it remembered at training time: dated frameworks, generic layouts, and missed accessibility, speed and security basics. - **What you get:** research instructions that make your AI look up what is current on the day you ask, plus design, coding, accessibility, performance, search and security playbooks and checklists it follows on every site. ## Quick start Run the npm installer in your project folder. It asks how you want to work, which host to use and where to save the work: ```bash npx website-build-skill ``` ![A short tour of what your AI learns, the 14 research domains and the 12 stages, ending on the install command on the closing card](https://raw.githubusercontent.com/aunysillyme/website-build-skill/main/docs/trailer.gif) It copies the skill pack into your host's folder, checks the saved files and writes a receipt. Open your AI and ask it to read the installed `SKILL.md`. Use `--dry-run` to preview paths. The [installer guide](docs/INSTALLER.md) covers all flags, receipts and exit codes. To remove it later, see [Uninstall](#uninstall). For a headless install with no prompts, name SOLO or TEAM and your host: ```bash npx website-build-skill --solo --target codex --dir . --yes npx website-build-skill --team --target claude-code --dir . --yes ``` ### Start with one paste For a chat AI, paste this starter and then describe the website you want: ```text Use option 1: Teach my AI. Help me build a website with Research, Learn, Ingest, Scope, Match, Mock, Choose, Build, Prove, Protect, Challenge and Ship. First act as Researcher: establish the request date from host evidence, then research the site's design, accessibility, performance, security, stack and search needs using current opened sources. Save a library with source URLs, publication dates or unknown, access dates, disagreements and checklists. When search is unavailable, report research as BLOCKED and return a query plan. If you lack files, return named bodies for me to save and reattach before passing the save gate. Only then confirm my brief, extract my brand or request profile/banner screenshots, and show three visual mockups. Wait for my named selection before site code. Choose the smallest fitting stack. Keep evidence and handoffs between roles. Protect means security headers and the ship security checklist; Challenge means a different model family reviews a fixed candidate read-only, and I authorize release. This starter is a compact entrypoint, so claim Ship only after the five bundles or the skill folder supply the full twelve-stage route and its checklists. Start with the site request I provide next; ask only for missing inputs that change the work. ``` For the complete method, paste or attach the five bundles from one revision: [method](docs/bundles/method.md), [prompts](docs/bundles/prompts.md), [playbooks](docs/bundles/playbooks.md), [checklists](docs/bundles/checklists.md), and [team](docs/bundles/team.md). Get them from [the public bundle folder](https://github.com/aunysillyme/website-build-skill/tree/main/docs/bundles) or [the main archive](https://github.com/aunysillyme/website-build-skill/archive/refs/heads/main.zip). Ask your AI to name the files it read before starting. ## What your AI learns | Area | What your AI learns | Files | |---|---|---| | Research | It learns to inspect current sources across fourteen domains, record source dates and evidence, and recheck stale claims. | [Research prompt](skills/website-build-skill/prompts/01-website-deep-research.md), [research and memory](skills/website-build-skill/playbooks/research-and-memory.md) | | Brand and design | It learns to extract brand tokens, distinguish observed rules from inferences, and compare three desktop and mobile designs through hierarchy, typography and composition. | [Brand extraction](skills/website-build-skill/prompts/03-extract-brand-kit.md), [three mockups](skills/website-build-skill/prompts/04-three-mockups.md) | | Graphics and image rights | It learns to verify intended-use rights, choose responsive crops and formats, and inspect exported dimensions, compression and alt-text intent. | [Image pipeline](skills/website-build-skill/playbooks/image-pipeline.md), [brand and mockup checklist](skills/website-build-skill/checklists/brand-and-mockups.md) | | Stacks and code | It learns to choose between static HTML, a scripted static builder and a framework, validate content records, and render shared templates safely. | [Stack selection](skills/website-build-skill/playbooks/choose-stack.md), [data and templates](skills/website-build-skill/playbooks/data-and-templates.md) | | Languages and code quality | It learns current HTML, CSS and JavaScript/TypeScript practice for its chosen stack, including tests, linting and secure code. | [Coding research](skills/website-build-skill/prompts/13-coding-research.md), [code-quality checklist](skills/website-build-skill/checklists/code-quality.md) | | Accessibility | It learns to use semantic structure, test keyboard and screen-reader journeys, compute contrast, and check reflow and reduced motion. | [Accessibility playbook](skills/website-build-skill/playbooks/accessibility.md), [accessibility checklist](skills/website-build-skill/checklists/accessibility.md) | | Performance | It learns to set route budgets, run repeatable Lighthouse measurements, distinguish lab scores from field data, and diagnose image, font and script costs. | [Performance budgets](skills/website-build-skill/playbooks/performance-budgets.md), [performance and search checklist](skills/website-build-skill/checklists/performance-seo.md) | | Search and answer engines | It learns SEO and AEO through visible answers, truthful structured data, canonical URLs, sitemaps and llms.txt generated from shared records. | [SEO and AEO prompt](skills/website-build-skill/prompts/08-seo-aeo.md), [structured data and llms.txt](skills/website-build-skill/playbooks/structured-data-and-llms.md) | | Security headers | It learns to tailor CSP and other headers to actual resources and response types, verify HTTPS readiness, and test journeys with the policy enforced. | [Security headers](skills/website-build-skill/playbooks/security-headers.md), [security checklist](skills/website-build-skill/checklists/security.md) | | Hosting and operations | It learns to promote an exact release artifact, verify live routes, plan rollback, and document monitoring, recovery and maintenance ownership. | [Deploy and operate](skills/website-build-skill/playbooks/deploy-and-operate.md), [ship and verify](skills/website-build-skill/prompts/11-ship-and-verify.md) | | Review and release | It learns to prepare a fixed candidate for review by a different model family, reproduce findings, verify fixes, and check release authority and evidence. | [Adversarial review](skills/website-build-skill/prompts/10-adversarial-pre-ship.md), [ship checklist](skills/website-build-skill/checklists/ship.md) | ## Part of a set Three open-source tools that work on their own and fit together: | Repo | What it gives you | |---|---| | [agent-personalizer](https://github.com/aunysillyme/agent-personalizer) | One interview writes the profile and rules every AI you use reads, kept in sync from one source. | | [model-orchestrator](https://github.com/aunysillyme/model-orchestrator) | Model router for AI coding agents: installs routing rules, 8 subagents, hooks and a CLI runner so your AI picks model and effort per task and saves tokens. | | **website-build-skill** | A skill pack that teaches your AI current website-building expertise: research, design, code, accessibility, performance, search and security. | ## How your AI uses it on a site Every site follows the same twelve stages, so your AI researches before it designs and checks before it ships. 1. **Research:** save current sources, domain coverage and a research gate receipt. 2. **Learn:** read the library and learn the checklists for the next job. 3. **Ingest:** apply that research and record what the role learned. 4. **Scope:** confirm the brief, available tools and artifact owners. 5. **Match:** turn brand assets into rules, tokens and approved inferences. 6. **Mock:** present three visual directions for your named selection. 7. **Choose:** choose the smallest stack that fits the approved direction. 8. **Build:** implement the selected design. 9. **Prove:** gather accessibility, performance and search evidence. 10. **Protect:** check security headers and exposed surfaces. 11. **Challenge:** have a different model family review the fixed candidate read-only. 12. **Ship:** authorize release, check the live site and save its operations document. [Method details](docs/METHOD.md) · [fictional brief](skills/website-build-skill/examples/fictional-studio-brief.md). ## Two ways to work | Choice | What happens | Works with | |---|---|---| | **1 · Teach my AI** (SOLO) | One AI learns the whole method and does every job itself. Nothing else to install. | Any AI: Claude, ChatGPT, Grok, Cursor, Copilot | | **2 · Give me the team** (TEAM) | Seven specialist agents, one per job. One command installs their files; you open each session yourself and carry the handoffs between them. Researcher goes first. | Claude Code, Codex, Hermes, Gemini (via Antigravity CLI), Grok Bot | Not sure? Pick 1. Moving to 2 later keeps everything you have already done. The installer also asks where to save the work: an Obsidian vault folder, a folder on this computer (default: your project folder), Notion (a local working copy plus an export of the finished library), or another local folder. That folder becomes the workspace root. SOLO follows the same jobs with saved self-handoffs. TEAM uses seven sessions you open yourself, with handoffs you carry between them and Researcher first: - **Researcher:** saves the shared library before the other roles work. - **Coordinator:** accepts evidence, owns the brief and routes each handoff. - **Designer:** extracts brand rules and presents the visual directions. - **Graphics:** produces assets and records rights, formats and exports. - **Builder:** implements the chosen direction and reproduces review findings. - **Optimizer:** checks search, answer-engine and discovery requirements. - **Reviewer:** reviews the fixed candidate read-only from a different model family. ## Works with whatever your AI has | Your AI has | You get | |---|---| | Web search | Research current to the date of your request | | File access | A research library saved in your project and read back | | Image tools | Three visual designs to choose from | | A second AI from another model family | An independent review before release | | A browser or terminal | Your AI runs speed, accessibility and live checks | | Separate agents | The seven-role team working in parallel | | Deploy access | Your AI ships the authorized site and checks it live | With any of these missing, the AI hands you the exact next step (a query plan, file contents to save, a review packet, commands to run) and the stage continues when you bring the result back. ## Choose your host Each guide below has a documented setup with its own loading check. Native loading is tested on Claude Code, Codex and the Grok Build CLI with dated receipts (one run each, limitations listed), and [compatibility](docs/COMPATIBILITY.md) tracks current status host by host. | Host | Set it up | What you get | |---|---|---| | [Claude Code](adapters/claude-code.md) | Install with `--target claude-code`; ask it to read `.claude/skills/website-build-skill/SKILL.md` | Method files; TEAM adds the full role bundles | | [Codex](adapters/codex.md) | Install with `--target codex`; ask it to read `.agents/skills/website-build-skill/SKILL.md` | Method files; TEAM adds the full role bundles | | [Hermes](adapters/hermes.md) | Install with `--target hermes`; follow the trust and loading steps | Shared method files; TEAM adds the full role bundles | | [Gemini (via Antigravity CLI)](adapters/antigravity.md) | Install with `--target antigravity`; read `.website-build-skill/SKILL.md` | Portable method files; TEAM adds the full role bundles | | [Grok Bot](adapters/grok.md) | Attach the five bundles and follow the manual team setup | Complete role instructions and addressed handoffs | | [Claude Desktop](adapters/claude-desktop.md) | Attach the five bundles and paste the shared instructions | The SOLO method in your chat | | [ChatGPT](adapters/chatgpt-project.md) | Add the instructions and five bundles to a Project or [custom GPT](adapters/custom-gpt.md) | The SOLO method and saved state packets | See [compatibility](docs/COMPATIBILITY.md) for tested hosts, documented setups and the evidence for each. For manual copying, preserve the complete skill folder, its manifest and its subdirectories in a new destination. ## Where your work is saved Choose an existing local directory during setup: a folder in your Obsidian vault, your website project or another local destination. That output root holds `site-work/`, including research, source dates, brand decisions, checklists and handoffs. Notion is an export destination with an authoritative local working copy. For headless setup, `--output-dir` records the existing output root; otherwise Researcher asks before writing. Reopen that library for later pages and refresh claims against the date of the new request. ## Tips - **Brand:** supply approved assets and review screenshot-derived guesses before they become rules. - **Choice:** compare hierarchy, density, typography and imagery across the mockups. - **Reuse:** reopen research and refresh stale claims before another page or role. - **Scope:** keep one primary action and choose the smallest fitting stack. - **Evidence:** save each measured result beside its command and candidate revision. [More tips](docs/TIPS.md) · [documentation index](docs/README.md). ## For agents Read [llms.txt](llms.txt) for the source catalog and [AGENTS.md](AGENTS.md) for repository instructions. For a headless project install, name SOLO or TEAM: ```bash npx website-build-skill --solo --target codex --dir . --yes npx website-build-skill --team --target claude-code --dir . --yes ``` Follow the installed `SKILL.md` and the recorded mode and output root in the receipt. The [installer contract](docs/INSTALLER.md) defines preflight checks, exit codes and receipt-based removal. ## Common questions **How do I teach Claude Code or Codex to build better websites?** Run `npx website-build-skill`, pick your host and let it copy the skill pack into your host's folder, then ask your AI to read the installed `SKILL.md`. It learns current practice in research, design, code, accessibility, performance, search and security before it touches your site. **Which AI tools does it work with?** Any AI that can read files works with the SOLO method, including Claude, ChatGPT, Grok, Cursor and Copilot. TEAM installs the seven workers' files for Claude Code, Codex, Hermes, Gemini (via Antigravity CLI) and Grok Bot in one command; you open each session yourself and carry the handoffs between them, with Researcher first; see [Choose your host](#choose-your-host) for each setup guide. **What does my AI research before it builds?** It researches fourteen domains current to the day of your request, from design and accessibility to performance, security, stack and search, and saves a library with source URLs and access dates so later pages can refresh stale claims. See [What your AI learns](#what-your-ai-learns) for the full list. **Can one AI run it, or does it need a team?** Either. SOLO has one AI learn the whole method and do every job itself with nothing else to install; TEAM installs the seven workers' files in one command; you open each session yourself and carry the handoffs between them. Researcher goes first. Pick SOLO if you are not sure; moving to TEAM later keeps everything you have already done. **How do I remove it?** Follow [Uninstall](#uninstall), using the target and directory from your install. It removes only files the installer wrote that still match its receipt. See the [installer guide](docs/INSTALLER.md) for flags and exit codes. ## Uninstall ```bash npx website-build-skill --uninstall --target codex --dir . --yes ``` Use the target and directory from your install; it removes only files the installer wrote that still match its receipt. ## Read next - [Installer guide](docs/INSTALLER.md) - [Method](docs/METHOD.md) - [Compatibility](docs/COMPATIBILITY.md) - [Tips](docs/TIPS.md) - [Evaluation](docs/EVALUATION.md) - [Provenance](docs/PROVENANCE.md) - [Documentation index](docs/README.md) ## Contributing, credits and license Contributions welcome: reproducible defects, clearer instructions, accessible examples and host evidence. Read [CONTRIBUTING](CONTRIBUTING.md), [the conduct policy](CODE_OF_CONDUCT.md), [SECURITY](SECURITY.md), [maintainers](MAINTAINERS.md) and [release operations](RELEASING.md). [Provenance](docs/PROVENANCE.md) records the method's sources and attribution. [Evaluation](docs/EVALUATION.md) describes the checks; run `npm run build`, `npm run check` and `npm test` locally. Original package work is licensed under [MIT](LICENSE); the covenant retains its attribution. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/guides/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/README.md Relative links below resolve against that repository source location. --- # Documentation Start with the guide for the job you want to do: | Guide | What it gives you | |---|---| | [Method](METHOD.md) | The stages, role owners and evidence gates | | [Installer](INSTALLER.md) | Setup commands, destinations, receipts, exit codes and uninstall | | [Compatibility](COMPATIBILITY.md) | Tested hosts, documented setups and evidence requirements | | [Tips](TIPS.md) | Brand, design, scope and research practices | | [Provenance](PROVENANCE.md) | Sources, attribution and the canonical file map | | [Evaluation](EVALUATION.md) | Repository checks and adversarial test cases | | [Manual host gate](evaluation/host-gate.md) | Live-host skips, per-OS loading and ZIP trials, with required receipts | | [Library check](LIBRARY-CHECK.md) | Validate saved research claims and source receipts | | [Bundles](#bundles) | The full method as files to attach or paste into a chat | ## Bundles Use files from one revision together: - [Method](bundles/method.md): entrypoint and research contract. - [Prompts](bundles/prompts.md): stage instructions. - [Playbooks](bundles/playbooks.md): implementation and verification guidance. - [Checklists](bundles/checklists.md): stage evidence and release checks. - [Team](bundles/team.md): complete roles and manual team setup. [Quick start](../README.md#quick-start) · [Agent source catalog](../llms.txt). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/installer/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/INSTALLER.md Relative links below resolve against that repository source location. --- # Installer contract `npx website-build-skill` copies the canonical files, verifies them by read-back and writes a receipt of exactly what it wrote. Open the chosen host and ask it to read the installed entrypoint. See [compatibility](COMPATIBILITY.md) for host setup evidence and loading checks. TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. ## Manual copy Copy the complete `skills/website-build-skill/` directory into a new, empty destination: | Target | Project destination | Loading step | | --- | --- | --- | | Claude Code | `.claude/skills/website-build-skill/` | Read back the entrypoint and assets | | Codex | `.agents/skills/website-build-skill/` | Inspect discovery and invocation | | Hermes | `.agents/skills/website-build-skill/` | Check trust and shared discovery | | Gemini (via Antigravity CLI) | `.website-build-skill/` | Ask the session to read the entrypoint | | Generic agent | `.website-build-skill/` | Read the entrypoint and preserve existing routers | | Grok file-capable host | `.grok/skills/website-build-skill/` | Read back the entrypoint, or paste the bundles | Do not merge into a conflicting directory. Compare and preserve existing files first. Copy a router pointer into existing project instructions only with authorization; do not replace them. Confirm the host can read the copied entrypoint and assets before beginning. For chat-only hosts, attach the five bundles and use the [Project instructions](../adapters/chatgpt-project.md). ## Command interface Install from npm: ```bash npx website-build-skill --solo --target codex --dir . --scope project --yes npx website-build-skill --team --target antigravity --dir . --scope project --yes ``` This recorded install uses the published 0.1.13 package (SOLO, Codex target), ending on the receipt read-back of 57 files. ![Installer demo: preview the files, install the Codex skill and inspect the receipt](https://raw.githubusercontent.com/aunysillyme/website-build-skill/main/docs/demo.gif) For unreleased changes, use `npx github:aunysillyme/website-build-skill` with the same flags. Append `#` to the repository reference for a reproducible install. - Choice: mutually exclusive `--solo` / `--team`; no arguments asks the canonical human question first. - Target: `claude-code`, `codex`, `hermes`, `antigravity`, `grok`, `agents`, `chatgpt`, `portable`. - Scope: project by default; user scope requires explicit selection, never implicit fallback. - Destination: `--dir` is explicit; `--bundle-dir` is accepted only for the Hermes team alias and resolves relative to the installed skill folder. Paths outside that folder are refused with exit 3 and "Path escapes destination root". - Output root: `--output-dir` selects an existing directory separately from the skill destination. Interactive setup collects the save-location choice and path; Enter on the local-folder path uses the website project directory. The receipt records `outputRoot` and optional `outputStorage` (kind and Notion export target). Researcher consumes those values before writing `/site-work/research/library/scope.md`. The installer only probes the root; it does not create a library or export to Notion. Headless setup without `--output-dir` leaves the root unset for Researcher to ask before writing. - Notion: export destination only. Interactive setup requires both the existing local working directory (authoritative) and a page URL or name for later export. No credentials are requested and no network upload occurs. - Writability: confirm the output root by writing and reading back before any sweep; an unwritable root fails without partial writes. - Headless: `--yes` requires mode, target and destination; EOF and missing arguments fail without writes. - Preflight: validate the entire intended output set before writes; reject symlinks, traversal and conflicts. - Paths: the project root you name is canonicalized, so a symlinked ancestor such as macOS `/tmp` installs normally. Every segment at or below that root is refused when it is a symlink, because that is what a third party can plant in an install target. - Idempotence: identical bytes are a no-op; changed bytes produce a reviewable diff without overwriting. - Dry run: `--dry-run` makes no writes, including directories, receipts or aliases. - Runtime: offline after package retrieval; no telemetry, profiles, background tasks or lifecycle hooks. - Receipt: list every actual path and digest, mode, target, definition and unresolved activation check. - Integration: preserve existing routers and emit a named integration snippet; unmerged means activation pending. - Interruption: record truthful partial results; never advertise an all-or-nothing transaction without evidence. - Exit policy: 0 complete, 2 invalid input, 3 preflight conflict, 4 interrupted write, 5 activation pending. 5 is a completed install whose integration snippet is unmerged, not a failure. - Uninstall: receipt-based, removing only files this package actually wrote and whose bytes still match the receipt. A file that was already on disk before the install, and therefore recorded as identical, is kept: installing it was a no-op and removing it would be taking something that is not ours. A changed file is kept and named, and so is a file the receipt does not list. It removes empty directories it created INSIDE the destination; the destination directory itself and any parent directories it had to create are left in place, because the receipt cannot tell them apart from directories that were already there. - Destination: decided by `--target` and `--dir`, or by where the receipt sits when only `--receipt` is given. It is never read out of the receipt's own text, so an edited or planted receipt cannot point a removal somewhere else. - Read-back: every file is re-read from disk and digested after the write pass. A digest taken from the bytes the installer meant to write is a record of the intention, not of what is on disk, and the receipt only ever carries the second one. - Ownership across runs: a reinstall that writes fewer files keeps owning what an earlier run wrote, so a SOLO install over a TEAM install does not orphan the bundles. ## TEAM sessions and handoffs The command copies the complete role files and five bundles. Open each worker's session yourself and supply its complete role body, readable resources and current handoff. Start Researcher first; carry its passing receipt and library to Coordinator before passing assignments to later workers. Record seven setup acknowledgements and use a different-family Reviewer. Enforce read-only review through host controls or a read-only packet. The Hermes alias copies bundles into the explicitly selected directory inside the skill folder. Session references and permissions require your own checks in the chosen host. See [Gemini (via Antigravity CLI)](../adapters/antigravity.md), [Hermes](../adapters/hermes.md) and [team ownership](../adapters/team.yaml). ## Verify the installation Run the local checks with `npm run check` and `npm test`. The installer tests cover traversal and symlink refusal, whole-set conflict refusal, idempotent reruns, dry-run behavior, output-root writability, receipt digests, changed-file preservation during uninstall and router integration. After installing, compare the receipt with files on disk and follow your adapter's loading check. Record client-trial evidence in [compatibility](COMPATIBILITY.md). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/compatibility/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/COMPATIBILITY.md Relative links below resolve against that repository source location. --- # Compatibility ## Tested hosts Three hosts have a dated native-loading receipt. Each covers SOLO skill discovery and loading on the exact version below, from one run under the operator's existing client configuration (each receipt lists its limitations); the full SOLO build route, TEAM and deployment stay documented until a trial records them. | Host | Version and system | Mode and scope | Date | Receipt | |---|---|---|---|---| | Claude Code | 2.1.281, macOS 26.6.2 arm64 | SOLO, native loading tested | 2026-09-24 | [receipt](evidence/host-trial-claude-code-2026-09-24.json) | | Codex CLI | 0.154.0, macOS 26.6.2 arm64 | SOLO, native loading tested | 2026-09-24 | [receipt](evidence/host-trial-codex-2026-09-24.json) | | Grok Build CLI | 1.0.40, macOS 26.6.2 arm64 | SOLO, native loading tested in a trusted folder | 2026-09-24 | [receipt](evidence/host-trial-grok-build-2026-09-24.json) | Every other host has a documented setup and a check you run, in its own adapter guide, to confirm it loaded on yours. A host moves from documented to tested when its receipt records the client/runtime version, trial date, operating system, package revision, installed paths and hashes, invocation, observed output, failure cases and remaining checks, kept sanitized in this repository. Hosted repository checks have a [revision-bound baseline receipt](evidence/audit-baseline.json). ## Hosts with a documented setup These guides describe file-copy, upload and manual-session routes. Their official documentation links are verification targets for a client trial. A documented setup becomes a tested-host entry when its receipt demonstrates loading, file access, tool permissions and the selected workflow. | Host | Setup guide | What to verify | Official reference | |---|---|---|---| | Claude Code | [Skill or plugin](../adapters/claude-code.md) | Discovery, invocation, assets and worker isolation | [Skills](https://code.claude.com/docs/en/skills) | | Codex | [Skill and explicit router](../adapters/codex.md) | Shared destination, role configuration and actual model family | [Skills](https://learn.chatgpt.com/docs/build-skills) | | Hermes | [Skill and separate sessions](../adapters/hermes.md) | Trust, alias semantics, routes and inherited permissions | [Skills](https://hermes-agent.nousresearch.com/docs/user-guide/features/skills) | | Gemini (via Antigravity CLI) | [Explicit core pointer](../adapters/antigravity.md) | Schema, paths, quoted off, routing and denied writes | [Subagents](https://antigravity.google/docs/subagents) | | Grok Bot | [Private skill and six-Bot group](../adapters/grok.md#grok-bot), from the xAI docs read 2026-09-25 | `/` lists the saved skill, role files in `/workspace`, group routes; Reviewer outside Grok Bot | [Bots](https://docs.x.ai/grok-bot/bots) | | Claude Desktop | [Bundle attachments](../adapters/claude-desktop.md) | Account upload controls and loaded resources | [Skills](https://support.claude.com/en/articles/12512180-use-skills-in-claude) | | ChatGPT Project/custom GPT | [Instructions and bundles](../adapters/chatgpt-project.md) | File access, tool availability and state reattachment | [Projects](https://help.openai.com/en/articles/10169521-projects-in-chatgpt) | | Generic file-reading agent | [Explicit core pointer](../adapters/agents-md.md) | Router read-back and preserved instructions | [Adapter](../adapters/agents-md.md) | ## Evidence to collect - **Documentation:** save the opened official source, version, access date and setup facts it supports. - **Native loading:** record the actual entrypoint and every resource the client can read. - **TEAM:** one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Record separate session references, setup acknowledgements, model families and permission probes. - **Delivery:** record fresh package installs for each operating system being claimed, including Linux and Windows. - **Uploads and URLs:** verify account ZIP controls and raw-URL retrieval in the actual client. - **Runtime:** the package declares Node 18 or newer; [the check workflow](https://github.com/aunysillyme/website-build-skill/blob/main/.github/workflows/check.yml) defines its runtime matrix. ## Working with the tools you have The [canonical tool table](../skills/website-build-skill/playbooks/research-and-memory.md#working-with-the-tools-you-have) supplies the next step for research, saved files, visual mockups, measurements, independent review and release. Each stage resumes when its required tool output or evidence is available. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/method/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/METHOD.md Relative links below resolve against that repository source location. --- # Method Research. Learn. Ingest. Execute: Match. Mock. Build. Prove. Ship. The [canonical skill](../skills/website-build-skill/SKILL.md) owns the method. | Stage | Owner | Reviewable result | | --- | --- | --- | | Research | Researcher | Saved library, fourteen-domain coverage, sources and gate receipt | | Learn and ingest | Every role | Reopened library, applied checklist and learning receipt | | Scope | Coordinator | Brief, capabilities and artifact ownership | | Match | Designer | Brand rules, tokens, conflicts and approved inferences | | Mock | Designer with Graphics | Three visual directions and a named human selection | | Choose and build | Builder | Smallest justified stack and selected implementation | | Prove | Builder, Designer, Optimizer | Accessibility, performance and discovery evidence | | Protect | Builder | Headers and exposed-surface checks | | Challenge | Different-family Reviewer | Read-only findings against a fixed candidate | | Ship | Coordinator then Builder | Authorized promotion, live checks and operations document | ASK DATE is the date of the user's request, fixed at engagement start. Researcher establishes it before intake using the [canonical rule](../skills/website-build-skill/playbooks/research-and-memory.md#ask-date-rule). A source publication/update date describes the source. An access date records an actual opening. Revalidation records whether a claim still applies and what was reopened. A freshness window limits reuse. A derived floor is a qualified lower bound, never proof of the actual day or later currency. Missing provenance or current research keeps dependent work BLOCKED. SOLO follows all seven roles sequentially with self-handoffs and a framing reset. TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Record seven setup acknowledgements, one writer per artifact, and a saved research gate. Neither mode earns independent review from a second name in the same model family. Use PASS, FAIL, BLOCKED or NOT APPLICABLE with evidence. Never convert absence into a pass. The user owns the resulting library, claim ledger, domain checklists, decisions and handoffs. Keep them with the website project. The skill repository is the method, not storage for client work. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/evaluation/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/EVALUATION.md Relative links below resolve against that repository source location. --- # Evaluation and evidence Local deterministic checks and live agent behavior are different claims. The checks below reject repository defects; they cannot prove a host obeys an instruction. The [baseline receipt](evidence/audit-baseline.json) records a hosted CI run for one exact revision. Platform activation, package publication and live model behavior remain unproved by that run; check the current candidate separately. ## Run the repository gates ```sh node scripts/check.mjs node --test test/*.test.mjs ``` Check performs no writes. The test suite copies synthetic fixtures into `.test-work/`, mutates one condition, asserts the named rejection and removes each scratch copy. The GREEN control and the RED cases run against the same policy the CLI and CI use; there is no separate test-only exemption. The public-identifier policy is resolved: the owning account is permitted on the exact CODEOWNERS line and inside repository URLs, and the supported host's product name is permitted as a product name. A path-shaped use of that product name is still rejected, so a local working directory cannot pass through the exemption, and a test covers that case. Private identifiers are also rejected in filenames, not only in file contents. | RED case | Named rejection | Regression | | --- | --- | --- | | Private machine path | PRIVACY | RED01 in privacy.test.mjs | | Private issue URL or identifier | PRIVACY | RED02 in privacy.test.mjs | | Unavailable install advertised without a same-line label | INSTALL_CLAIM | RED03 in content.test.mjs | | Missing Markdown relative target | LINK | RED04 plus image/reference/HTML/anchor controls | | Generated bundle bytes drift | BUNDLE_DRIFT | RED05 in content.test.mjs | | Floating action reference | ACTION_PIN | RED06 plus reusable-action control | | Write token on a read-only job | WORKFLOW_PERMISSIONS | RED07 plus top-level/all-write controls | | Inactive archive enters active prompts | ARCHIVE_EXCLUSION | RED08 plus manifest route checks | | Unsupported test, measurement, publication or compatibility claim | CLAIM_EVIDENCE | RED09 in README, new docs and CHANGELOG | ## Editorial claim gate Every public Markdown document outside canonical assets, generated bundles and inactive test fixtures has a reviewed checksum in [content-review.json](evidence/content-review.json). New or changed prose fails closed until a maintainer reviews it. This is intentionally broader than a keyword search. The generator never updates that ledger. Canonical claims remain covered by manifest hashes and source review; generated text must exactly match those sources. A checksum binds review scope, not factual truth. The current disposition permits instructions and explicitly qualified status only. A positive claim instead needs its exact text and a JSON evidence receipt under `docs/evidence/` with command, environment, date, exitCode, artifactSha256 and observed result. Reviewers must read that receipt and reproduce material results; a structurally valid receipt can still be false. A change to the ledger itself requires code-owner review. No content checker can replace that authority. ## Privacy scope The gate scans every public file, including hidden community metadata, fixtures and generated bundles. It rejects personal paths, private identifiers, personal names and email addresses without echoing their values. Every exemption is narrow, exact and implemented once in `privacyIssues` (`src/validate.mjs`). The privacy tests run against that same function. - License: the one exact copyright line in `LICENSE`. - CODEOWNERS: the one exact owner line in `.github/CODEOWNERS`. - Product name: the supported host's public product name, as a name. Path-shaped uses stay rejected. - Own repository: `https://github.com/aunysillyme/website-build-skill` and the matching `raw.githubusercontent.com` URL, plus the npm shorthand `github:aunysillyme/website-build-skill`. - Funding: exactly `https://github.com/sponsors/aunysillyme`. Sentence punctuation after it (`. , ; : ] > *`) is allowed. A different handle, a longer path, a query or a fragment fails. - Sibling repositories: exactly `https://github.com/aunysillyme/agent-personalizer` and `https://github.com/aunysillyme/model-orchestrator`, in GitHub URL, raw file URL and npm shorthand spellings. A third repository or a longer name fails. The tests reject a third repository, longer names, other Sponsors handles and adjacent private text. These exemptions do not permit personal identifiers anywhere else. The unchanged inactive archive is checksum-pinned; its historical path is never copied into active prompts. Scratch fixture copies and version-control internals are outside the public-tree walker. The final raw recursive sweep is a separate report step, so any literal policy conflict stays visible. ## Structural checks Manifest assets must exist, match digests and contain all seven populated role bodies. The README question must match the canonical template. Every generated bundle uses manifest order. The workflow validator accepts the JSON subset of YAML, checks every action reference recursively, requires read-only permissions except the release publish job's scoped OIDC token, disables persisted checkout credentials, and requires concurrency cancellation and job timeouts. Other YAML syntax fails closed pending parser review. External URL reachability is outside the offline link check. ## Planned live behavior evaluation Run the [end-to-end host trial](evaluation/host-trial.md) to collect a complete synthetic SOLO local-preview result and the negative cases below. Its receipt template is deliberately NOT RUN. The method's outcome remains unproved until actual artifacts and a reproducible trial are reviewed; adding the protocol does not establish support. | Scenario | Required observation | | --- | --- | | Missing brand kit | Requests suitable screenshots and labels inferences | | No search or stale library | Returns queries/unresolved claims; no downstream design | | Missing source or date | Blocks research and dependent roles | | SOLO transition | Reopens library, completes learning and saves self-handoff | | TEAM transition | Seven actual routes; Researcher first; Coordinator reads saved evidence | | Same-family Reviewer | INTERNAL critique, independent gate BLOCKED | | No memory | Returns a save/reattach packet without claiming persistence | | Conflicting accents | Surfaces conflict and requests only the decision that changes the design | | Hostile fetched text | Refuses instruction injection and keeps authorized write scope | | Reviewer allegation | Reproduces before recording a confirmed defect | | Date ladder and reuse | Rejects every inactive ASK DATE fixture and passes genuine evidence controls | These are evaluation specifications, not completed trials. Installer regressions run as ordinary tests. Cross-platform fresh-package and live-date evaluations remain TODO. Run live evaluations only with explicit tool access and an approved budget. ## File and evidence index - [Historical local run](evidence/local-checks.json): an earlier TypeScript-era source snapshot; preserve it as historical evidence, not current-candidate validation. - [Audit baseline](evidence/audit-baseline.json): exact baseline revision, local checks, hosted CI and registry observation. Rerun gates after changes. - [Floor coverage register](evidence/floor.json): exhaustive section-4 filenames and implementation scope. - [Content regressions](https://github.com/aunysillyme/website-build-skill/blob/main/test/content.test.mjs), [privacy regressions](https://github.com/aunysillyme/website-build-skill/blob/main/test/privacy.test.mjs). - [Installer boundary](https://github.com/aunysillyme/website-build-skill/blob/main/test/installer.test.mjs), [package boundary](https://github.com/aunysillyme/website-build-skill/blob/main/test/package.test.mjs), [date routing](https://github.com/aunysillyme/website-build-skill/blob/main/test/research-date.test.mjs). - [Synthetic missing asset](https://github.com/aunysillyme/website-build-skill/blob/main/test/fixtures/invalid-manifest.json), [brand conflict](https://github.com/aunysillyme/website-build-skill/blob/main/test/fixtures/conflicting-brand.json). - [Malicious source](https://github.com/aunysillyme/website-build-skill/blob/main/test/fixtures/malicious-source.md), [incomplete handoff](https://github.com/aunysillyme/website-build-skill/blob/main/test/fixtures/incomplete-handoff.yaml), [date cases](https://github.com/aunysillyme/website-build-skill/blob/main/test/fixtures/ask-date-cases.md). - [Bundle generator](../src/bundle.mjs), [validator](../src/validate.mjs), [build entry](https://github.com/aunysillyme/website-build-skill/blob/main/scripts/build.mjs), [check entry](https://github.com/aunysillyme/website-build-skill/blob/main/scripts/check.mjs). - [CLI contract](../src/cli.mjs), [choice accessor](../src/choice.mjs), [destination catalog](../src/catalog.mjs), [install contract](../src/install.mjs), [agent contract](../src/agents.mjs). - [Entry point](../bin/website-build-skill.mjs), [install engine](../src/install.mjs), [package metadata](../package.json), [lock metadata](https://github.com/aunysillyme/website-build-skill/blob/main/package-lock.json). - [Check workflow](https://github.com/aunysillyme/website-build-skill/blob/main/.github/workflows/check.yml), [release boundary](https://github.com/aunysillyme/website-build-skill/blob/main/.github/workflows/release.yml). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/provenance/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/PROVENANCE.md Relative links below resolve against that repository source location. --- # Provenance The method's research prompts originate in two public posts by the repository's author: - [Research, learn, ingest, then execute](https://x.com/i/status/2095267280053735464): graphics research and evidence-led site review. - [Current design research and three brand-matched mockups](https://x.com/i/status/2097418846282145946): browser exploration, brand inputs and design selection before code. Attribution comes from the supplied source record; live post retrieval is UNVERIFIED here. The [inactive graphics original](../skills/website-build-skill/archive/graphics-design-original.md) retains its bytes and historical save instruction. It is provenance reading only and is excluded from active bundles and stage routes. The website research and site audit prompts are adaptations. They are not a recovered historical audit transcript. Reported historical measurements are not repeated as package evidence. Original package work uses [MIT](../LICENSE). The conduct text retains its own attribution and license. No third-party site screenshots, fonts or brand kits are included. ## Design review techniques Two techniques adapted (own wording, not copied text) into checklists/brand-and-mockups.md (BM05), checklists/code-quality.md (CQ06), checklists/build.md (B08) and prompts/04-three-mockups.md: - Design Read (recording page type, audience and intended visual direction before judging a design) and contextual landing-page layout defaults: adapted from Taste Skill, [github.com/Leonxlnx/taste-skill](https://github.com/Leonxlnx/taste-skill), commit `ce26fc25c0e5e8cab638f883de62d9a86ee5e45b`, MIT licence. - A named template-tell check (signs of generic AI-template output) and separating measurable findings from visual judgment with in-context verification: adapted from Impeccable, [github.com/pbakaus/impeccable](https://github.com/pbakaus/impeccable), commit `9d715cc4f5564a990ca8345abfdd5df6dc9b41c8`, Apache-2.0 licence. ## Reference ledger These links are research entrypoints, not evidence that this package was loaded in a client. Open current sources and record real source/access dates before relying on a claim. - [WCAG contrast](https://www.w3.org/WAI/WCAG22/Understanding/contrast-minimum.html). - [Core Web Vitals](https://web.dev/articles/defining-core-web-vitals-thresholds). - [Search and AI features](https://developers.google.com/search/docs/appearance/ai-features). - [Browser examples](https://threejs.org/examples/). - [Host sources and settling checks](COMPATIBILITY.md). ## Automation pin provenance The checkout v7.0.1 tag points to commit `3d3c42e5aac5ba805825da76410c181273ba90b1`. Source: [upstream tag](https://github.com/actions/checkout/releases/tag/v7.0.1), checked through the GitHub API in the baseline receipt. This identifies the pin; it is not an audit of the action or a hosted workflow run. The setup-node v7.0.0 tag points to commit `820762786026740c76f36085b0efc47a31fe5020`. Source: [upstream tag](https://github.com/actions/setup-node/releases/tag/v7.0.0), checked through the GitHub API in the baseline receipt. These tag checks identify pins only; they are not an audit of the actions. See the [baseline receipt](evidence/audit-baseline.json) for the separate hosted-check observation. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/library-check/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/docs/LIBRARY-CHECK.md Relative links below resolve against that repository source location. --- # Library receipt checker ## What and why `check-library` checks that a research ledger's opened claims have consistent local fetch receipts. It cannot prove a fetch happened, that an excerpt is verbatim, or that the excerpt supports the claim. Independent live-page review remains required. ## Trigger Run before research acceptance and again when claims or receipts change. AD03, AD04 and AD07 consume the result; AD09 adds independent source re-fetching. ## Invocation chain Run `website-build-skill check-library `. In a repository checkout, run `node bin/website-build-skill.mjs check-library `. The binary calls `src/cli.mjs`, which calls `checkLibrary` in `src/library-check.mjs`. The binary prints the result and sets its process exit code. Exit 0 means local receipt consistency; exit 2 means invalid input or a failed check, printed to stderr. ## Dependencies Node and the package's own modules. No external packages, network access, fetch tool credentials, installation writes or background services are needed by the checker. ## Reads Relative to the supplied site-work directory, read `research/library/sources.md` and `research/library/receipts/.md` for each ledger claim. Refuse symlink paths and unsafe claim IDs. Claim IDs begin with `CLAIM-` and continue with letters, digits, hyphens or underscores. Every table row has these columns: ```text | Claim ID | Domain | Statement | URL | Published | Accessed | Status | Dependents | | --- | --- | --- | --- | --- | --- | --- | --- | ``` Use plain cells, escape a literal pipe as `\|`, and write `unknown` for an absent publication date. Column names ignore case, spaces, hyphens and underscores. The status vocabulary is `OPENED` or `UNVERIFIED`; legacy `Verified` labels in any case receive the same receipt checks as `OPENED`, never an exemption. Each receipt contains plain `url:`, `fetched_at:`, `tool:`, `result:`, `published:` and `excerpt:` fields. Use an environment-supplied UTC timestamp ending in `Z`, with seconds and optional three-digit milliseconds. Copy the publication date verbatim into both ledger and receipt, or use `unknown`. The date must appear in the excerpt. Put the verbatim supporting passage after `excerpt: |`, with every line indented two spaces; a single-line `excerpt:` is also accepted. The trimmed passage must contain at least 80 characters. Preserve fetch evidence at fetch time. ## Writes None. The caller captures the command, exit code and complete output in the gate record. The checker lists UNVERIFIED claims even on a successful consistency check. ## The closed loop Nothing watches a project library automatically. Researcher runs the check and corrects findings; Coordinator reopens the artifacts and records acceptance. An acted-on UNVERIFIED claim keeps acceptance BLOCKED even when the checker exits 0. Reviewer independently re-fetches the AD09 sample and reports excerpt mismatches. Send checker defects to the package maintainer with a synthetic reproducer; send source discrepancies to Researcher by claim ID and pause dependent decisions. ## Failure modes Missing or malformed ledgers, unsafe paths, duplicate claim IDs, absent or incomplete receipts, short excerpts, URL differences, inconsistent publication dates and blanket verification over unreceipted rows produce failures. A consistent fabricated receipt can still pass: this is a local consistency check, not fetch authentication. Timestamp provenance, live support for the claim and freshness require gate review. ## Run and verify by hand From the repository root: ```sh node bin/website-build-skill.mjs check-library test/fixtures/library-red node bin/website-build-skill.mjs check-library test/fixtures/library-green node --test test/library-check.test.mjs ``` Expect the RED fixture to exit 2 and name both unreceipted claims plus its blanket verification claim. Expect GREEN to exit 0 with counts. The fixtures are synthetic or inactive regression data, never live research evidence. ## Source of truth The implementation is `src/library-check.mjs`. The research policy lives in `skills/website-build-skill/checklists/research.md`; receipt authoring is specified in `skills/website-build-skill/templates/evidence.md` and `skills/website-build-skill/playbooks/research-and-memory.md`. Tests live in `test/library-check.test.mjs`. Hosted and live-client behavior remains UNVERIFIED until an actual trial is separately authorized and recorded. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/changelog/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/CHANGELOG.md Relative links below resolve against that repository source location. --- # Changelog All notable changes will be documented here. Format: [Keep a Changelog](https://keepachangelog.com/en/2.0.0/). Version policy: [Semantic Versioning 2.0.0](https://semver.org/spec/v2.0.0.html). ## [Unreleased] ## [0.1.15] - 2026-10-02 ### Changed - LICENSE names Auny LLC as copyright holder; outside contributions now require the CLA in CLA.md. ## [0.1.14] - 2026-10-01 ### Changed - Terminal demo re-recorded from the published 0.1.13 package. ## [0.1.13] - 2026-10-01 ### Added - README hero trailer: a short tour of what your AI learns, the 14 research domains and the 12 stages, ending on the install command. ### Changed - Terminal demo moved to the installer guide and labeled as a recording from version 0.1.1. - GIFs excluded from the npm package; documentation uses absolute raw image URLs. - The package check requires the GIF exclusion after the docs entry. - The privacy and voice scan reads a GIF by path only, so compressed bytes cannot match as prose. ## [0.1.12] - 2026-10-01 ### Added - Package funding metadata links to GitHub Sponsors, with a privacy exception for the exact URL and rejection coverage for other handles and paths. - README Uninstall and Read next sections, plus sibling repository links and descriptions in the generated `llms.txt` index. - Release completion checks cover the registry README, description, provenance attestations and funding metadata. ### Changed - Release operations name 57 installed files plus a receipt, the GitHub release creation command and registry lag handling. - Private vulnerability reporting text describes the enabled private advisory route. - Reader-facing host labels use Gemini (via Antigravity CLI), and the README places the related repository set after the skills overview. - The changelog format reference links to Keep a Changelog 2.0.0. - SECURITY.md describes the released installer, its guards and the supported version window. - REPO_SETTINGS.md records the default GitHub preview card as the decision. - Release operations create the GitHub release after the registry checks, compare the README to the tagged copy, and add funding, bin and demo image checks. - The README places Uninstall after Common questions, and the privacy gate documentation lists every public-identifier exemption. - The Sponsors URL exemption tolerates trailing sentence punctuation. ### Removed - The committed custom social preview image. - Stale release wording and the duplicate installer write description. ## [0.1.11] - 2026-09-30 ### Changed - CLAUDE.md now imports AGENTS.md so Claude Code loads the same method and contributor rules as every other agent. - The accuracy notes moved into AGENTS.md so every agent reads them. - AGENTS.md names `node scripts/build.mjs --rehash` for canonical asset edits. ## [0.1.10] - 2026-09-26 ### Added - `node scripts/build.mjs --rehash [ ...]` refreshes the manifest SHA-256 for only the named canonical assets and prints each old and new digest. It refuses the manifest itself, the archive and unknown paths, and writes nothing on refusal. Plain `node scripts/build.mjs` never changes manifest digests, so an unreviewed edit still fails the source-hash check. ### Changed - CONTRIBUTING and RELEASING name the `--rehash` step for intentional canonical edits. ## [0.1.9] - 2026-09-26 ### Fixed - `llms.txt` labels: the manifest entry names what it indexes instead of the bare word "manifest", and a label no longer ends in a colon. A test fails if any catalog label ends in a colon or is a bare asset kind. ## [0.1.8] - 2026-09-26 ### Added - Web design review in the existing checks. BM05 now names common template tells (gradient heading text, side-stripe card borders, the generic purple-to-blue palette, nested cards, icon-tile stacks, bounce easing, decorative numbered labels, thin border with a wide soft shadow) and contextual landing-page layout defaults. A recorded brand decision overrides each default, and text expansion, zoom and reflow always win. - CQ06 checks design-token drift and repeated copy-paste shortcuts, and asks that findings be verified in source before they are reported. B08 replays the chosen layout defaults on the built page at the recorded mobile and desktop viewports. - Prompt 04 opens with a one-line design read (page type, audience, direction) and points to BM05. - `docs/PROVENANCE.md` credits the techniques to Impeccable (Apache-2.0) and Taste Skill (MIT). No upstream text was copied and no dependency was added. ## [0.1.7] - 2026-09-25 ### Changed - Grok Bot has its own setup in the Grok adapter, written from xAI's Grok Bot documentation (read 2026-09-25) and the operator's own use: install the package into `/workspace`, save it as a private skill, and run TEAM as a six-Bot group with Reviewer outside Grok Bot, since every Bot runs on Grok. Grok Bot stays a documented setup with no trial receipt ([#15](https://github.com/aunysillyme/website-build-skill/issues/15)). - The shared manual fallback no longer names an instruction field or a seven-Bot group; it points to where each host keeps standing instructions and respects the host's group limit. - One description everywhere: `package.json`, the GitHub About box, npm and `.claude-plugin/plugin.json` now carry the same line, which matches the README lead and names the solo or team choice. The plugin manifest also gains the homepage and keywords. ## [0.1.6] - 2026-09-24 ### Added - Dated native-loading receipt for the Grok Build CLI 1.0.40 (SOLO, macOS, trusted folder, 2026-09-24) in `docs/evidence/`, linked from compatibility and the Grok adapter ([#15](https://github.com/aunysillyme/website-build-skill/issues/15)). ### Changed - The Grok adapter says to trust the project folder: Grok Build offers project skills only in a trusted folder, found during the trial. Grok Bot (the Bots app) is named separately as still awaiting its trial. ## [0.1.5] - 2026-09-24 ### Added - Packed consumer-install checks for SOLO and TEAM on a Linux, macOS and Windows CI matrix, plus explicit live-host skips and a manual receipt gate ([#16](https://github.com/aunysillyme/website-build-skill/issues/16)). - Dated native-loading receipts for Claude Code 2.1.281 and Codex CLI 0.154.0 (SOLO, macOS, 2026-09-24) in `docs/evidence/`, linked from compatibility and both adapter pages ([#15](https://github.com/aunysillyme/website-build-skill/issues/15)). ### Changed - TEAM consistently describes installing seven workers' files, opening each session yourself and carrying handoffs, with Researcher first; presentation checks guard that wording ([#17](https://github.com/aunysillyme/website-build-skill/issues/17)). ### Fixed - Receipts under a symlinked parent such as macOS `/tmp` are accepted; found during the 2026-09-24 host trials for [#15](https://github.com/aunysillyme/website-build-skill/issues/15). ## [0.1.4] - 2026-09-23 ### Added - Fourteenth research domain covering languages and coding practice, with dated primary-source research and a saved library checklist. - Coding-research prompt that turns the selected stack and versions into project coding standards before implementation. - Code-quality checklist wired into Builder and Reviewer gates, covering tests, linting, formatting, type-checking, dependencies and secure code. - Regression coverage for fourteen-domain research and freshness tables, and coding-research manifest and build links. ## [0.1.3] - 2026-09-23 ### Fixed - #13: Synchronized the repository settings description with `package.json`, added a description drift check and documented the release update step. - #14: Documented the installed skill folder as the Hermes alias path root in the adapter, installer contract and CLI help, with regression coverage for help text and alias placement. ## [0.1.2] - 2026-09-23 ### Changed - README "Two ways to work" is a two-row table in place of the raw installer prompt; the save-location choices are one sentence. SKILL.md keeps the card verbatim because the agent shows it to the user, and the drift check now requires it there only. ## [0.1.1] - 2026-09-23 ### Added - Documentation index, related tools table and agent quick-start links. - Installer demo recording script with receipt read-back and scratch cleanup. - Regression coverage for sibling privacy exceptions, role boundaries and demo failures. - `README.md` Common questions section answering what the package installs, which AI tools it works with, what it researches, SOLO vs TEAM, and how to remove it. - Headless TEAM install command alongside the existing SOLO example in the README Quick start and For agents sections. - `llms.txt` link labels generated from each source file's own heading or opening line, so a reader sees what a file teaches instead of a one-word kind. ### Changed - Reframed the package as a skill pack that teaches an AI current website-building expertise, with the twelve stages showing how it applies those skills. - README leads with the research-first method, setup and stage outcomes. - Host setup guides link to a shared compatibility and evidence reference. - Tool fallback tables, role boundaries and stack guidance lead with the next action. - Discovery index puts quick start, installer and compatibility ahead of source listings. - Package description summarizes the research, brand, design and build workflow. - Generated bundles carry the updated canonical instructions. - Every adapter's "Check that it loaded" section and the README host table now open with one positive sentence: native loading on that host is awaiting a real-client trial, and the check that follows confirms it on yours. `docs/COMPATIBILITY.md` states the same fact once, plainly, alongside what evidence moves a host from documented to tested. - Quick start reads "Run the npm installer in your project folder" instead of assuming a website project already exists. - The paste starter and the SOLO/TEAM choice card use affirmative phrasing with the same meaning: reporting research as BLOCKED when search is unavailable, claiming Ship only after the full route is available, and moving to TEAM later keeping everything already done. ### Fixed - Research gate reaches a mechanical checker from a copied install: PATH, then npx at the manifest version, then the recorded manual equivalent. - Restored the font-licensing verification trap on the Designer role, the role/permission-boundary fact on `SKILL.md`, and the four Grok Bot manual-relay clarifiers, all of which the 0.1.1 positive-framing rewrite had deleted instead of reframing (ROUND 1 audit, HIGH/MEDIUM/LOW). - `src/validate.mjs` sibling-repository privacy exemption now allows the same sentence-ending period as the self-repository exemption, without exempting a different repository name appended with a period (such as `.private`). ## [0.1.0] - 2026-09-23 ### Added - Per-claim page fetch receipts with verbatim excerpts and OPENED or UNVERIFIED ledger status. - Deterministic `check-library` CLI checks and independent live source sampling in the research gate. - Research-first website method, stage prompts, roles, playbooks and artifact templates. - Manual adapters, generated upload bundles and manifest-derived discovery index. - Community documents, repository settings checklist and local content gates. ### Changed - Research acceptance stays BLOCKED for acted-on claims without consistent fetch receipts. - Scope saves consistently use `/site-work/research/library/scope.md`. - Mockups require an image per direction and viewport, optionally with a static HTML preview. - Same-day receipt reuse follows the UTC calendar day within one engagement, with new fetches after day changes or change triggers. - First-run retrieval paths name the public repository tree, archive and raw files instead of waiting on publication. - The install choice card distinguishes file installation from manual worker setup; the installer does not start workers. - The zero-install starter names the full twelve-stage route, including Scope, Choose, Protect and Challenge. - Maintainer scripts use plain JavaScript ES modules without the former TypeScript runtime flag. - Repository settings are recorded as a dated live read-back rather than an unapplied proposal. - The installer the contract specified: preflight, symlink and traversal refusal, no-clobber idempotence, dry run, a receipt of per-file digests, receipt-based uninstall and exit codes. - Publishable package metadata with a real `bin`, an engine floor continuous integration runs, and a `v*` tag that publishes with provenance through npm trusted publishing. - Sources renamed from `.ts` to `.mjs`, since none of them contained TypeScript. The `--experimental-strip-types` flag, its runtime guard and the Node 22.6 floor are gone. ### Planned - Generated native workers for the seven-role team, gated on schema and permission tests. - Clean-client trials, behavioral evaluations and fresh-tarball trials on Linux and Windows. [Unreleased]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.15...HEAD [0.1.15]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.14...v0.1.15 [0.1.14]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.13...v0.1.14 [0.1.13]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.12...v0.1.13 [0.1.12]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.11...v0.1.12 [0.1.11]: https://github.com/aunysillyme/website-build-skill/compare/v0.1.10...v0.1.11 --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/codex/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/codex.md Relative links below resolve against that repository source location. --- # Codex TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. - Manual SOLO: copy the full canonical folder to a new `.agents/skills/website-build-skill/` in the chosen project. - Load the skill: ask the host to load `website-build-skill`; confirm the exact file it opens. - Install the files: `npx website-build-skill --team --target codex --dir . --scope project --yes`. - TEAM sessions use [the full team bundle](../docs/bundles/team.md). Do not overwrite `.codex/config.toml` or infer independence from two model names. - Shared core: Hermes may use the same candidate path; preserve identical content and retain the core until every owner releases it. - Sources to verify: [skills](https://learn.chatgpt.com/docs/build-skills), [custom agents](https://learn.chatgpt.com/docs/agent-configuration/subagents). ## Check that it loaded Native loading was tested on Codex CLI 0.154.0 (macOS, SOLO, 2026-09-24; one run, limitations in the [receipt](../docs/evidence/host-trial-codex-2026-09-24.json)); the check below confirms it on yours. Current status: [compatibility](../docs/COMPATIBILITY.md). Give the host a synthetic brief and ask it to identify the entrypoint and complete asset list. Record client version, date, exact path, tool availability and the observed invocation. Start Researcher, verify saved library read-back, then confirm Coordinator waits for its gate. For TEAM, record seven separate routes and families; an inherited family leaves independent review BLOCKED. Never grant a Reviewer write access just to persist findings. Return text for Coordinator to capture. When a tool is absent, keep current research BLOCKED until opened sources return; save and reattach named file bodies before the save gate; return inspected images before the visual gate. Read the [method](../docs/METHOD.md), [compatibility](../docs/COMPATIBILITY.md) and [installer contract](../docs/INSTALLER.md). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/claude-code/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/claude-code.md Relative links below resolve against that repository source location. --- # Claude Code TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. - Manual SOLO: copy the intact canonical folder into a new project `.claude/skills/website-build-skill/` directory. Preserve conflicts. - After checking native discovery, try `/website-build-skill`; for explicit loading, ask it to read its SKILL.md. - Plugin: the repository root contains metadata and discovers the canonical `skills/` tree. The plugin loads the method files. - For the plugin route, run the setup commands below and perform the loading check. - Plugin setup: `claude plugin marketplace add https://github.com/aunysillyme/website-build-skill`. - Plugin setup: `claude plugin install website-build-skill@website-build-skill`. - Check the plugin invocation: `/website-build-skill:website-build-skill`. - Install the files: `npx website-build-skill --team --target claude-code --dir . --scope project --yes`. - TEAM setup: open separate sessions with complete [role bodies](../docs/bundles/team.md), Researcher first. Use the [manual relay](grok.md#manual-fallback). - Source targets: [skills](https://code.claude.com/docs/en/skills) and [subagents](https://code.claude.com/docs/en/sub-agents). ## Check that it loaded Native loading was tested on Claude Code 2.1.281 (macOS, SOLO, 2026-09-24; one run, limitations in the [receipt](../docs/evidence/host-trial-claude-code-2026-09-24.json)); the check below confirms it on yours. Current status: [compatibility](../docs/COMPATIBILITY.md). Give the host a synthetic brief and ask it to identify the entrypoint and complete asset list. Record client version, date, exact path, tool availability and the observed invocation. Start Researcher, verify saved library read-back, then confirm Coordinator waits for its gate. For TEAM, record seven separate routes and families; an inherited family leaves independent review BLOCKED. Never grant a Reviewer write access just to persist findings. Return text for Coordinator to capture. When a tool is absent, keep current research BLOCKED until opened sources return; save and reattach named file bodies before the save gate; return inspected images before the visual gate. Read the [method](../docs/METHOD.md), [compatibility](../docs/COMPATIBILITY.md) and [installer contract](../docs/INSTALLER.md). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/hermes/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/hermes.md Relative links below resolve against that repository source location. --- # Hermes TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Check project trust and the installed client's loading controls before starting. Manual SOLO: copy the complete canonical folder into a new `.agents/skills/website-build-skill/` in the chosen project. For explicit loading, ask the session to read its SKILL.md directly. After confirming the installed CLI supports it, try preload: `hermes -s website-build-skill` after discovery is confirmed. After checking the installed CLI help, use its trust operation: `hermes skills trust`; this changes trust state and requires the user's authorization. Do not create version-control state or alter global settings during a copy. Install the files: `npx website-build-skill --team --target hermes --dir . --scope project --yes --bundle-dir team-aliases`. The path is relative to the installed skill folder, so the alias files land in `.agents/skills/website-build-skill/team-aliases/` and the path must stay inside that folder. ## Open the workers Open seven separate sessions yourself, Researcher first. Supply each complete role body from `.agents/skills/website-build-skill/roles/`, the linked resources and its current handoff. The `team-aliases` directory holds copies of the five bundles. An alias loads instructions; verify each worker through its own session reference and carry the handoffs yourself. Use [the team relay](grok.md#manual-fallback) and check each worker can read the assigned revision, including any uncommitted files it needs. Check Reviewer permissions on inherited tools; use a read-only packet and returned findings when needed. ## Shared core and receipt Codex and Hermes candidate paths overlap. Compare content before writing and preserve conflicts. Test installation in both orders, partial receipts and reference ownership before supporting shared uninstall. An explicit bundle directory authorizes only that destination. Never write a home-directory alias implicitly. The receipt enumerates core files, bundles, alias copies, hashes, mode and pending integration. ## Settling checks Verify project trust before/after explicit authorization, native preload, all manifest assets and each session reference. Verify seven acknowledgements, source-preserving relay, no downstream work before research and denied Reviewer writes. Probe image, browser, execution and independent-family access separately. Source targets: [skills](https://hermes-agent.nousresearch.com/docs/user-guide/features/skills), [delegation](https://hermes-agent.nousresearch.com/docs/user-guide/features/delegation), [CLI](https://hermes-agent.nousresearch.com/docs/user-guide/cli). ## Check that it loaded Native loading on this host is awaiting a real-client trial; the check below confirms it on yours. Current status: [compatibility](../docs/COMPATIBILITY.md). Give the host a synthetic brief and ask it to identify the entrypoint and complete asset list. Record client version, date, exact path, tool availability and the observed invocation. Start Researcher, verify saved library read-back, then confirm Coordinator waits for its gate. For TEAM, record seven separate routes and families; an inherited family leaves independent review BLOCKED. Never grant a Reviewer write access just to persist findings. Return text for Coordinator to capture. When a tool is absent, keep current research BLOCKED until opened sources return; save and reattach named file bodies before the save gate; return inspected images before the visual gate. Read the [method](../docs/METHOD.md), [compatibility](../docs/COMPATIBILITY.md) and [installer contract](../docs/INSTALLER.md). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/antigravity/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/antigravity.md Relative links below resolve against that repository source location. --- # Gemini (via Antigravity CLI) TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Check the installed client's file access and session permissions before starting. Manual SOLO: copy the canonical folder into `.website-build-skill/` and explicitly ask the session to read its SKILL.md. Install the files: `npx website-build-skill --team --target antigravity --dir . --scope project --yes`. ## Open the workers Open seven sessions yourself, Researcher first. In each session, load the complete `roles/.md` from `.website-build-skill/`, plus the linked resources and current handoff. Use the [manual relay](grok.md#manual-fallback) to carry packets and attachments. Record each session reference, actual model family and available tools. Keep command execution off by default; verify the host setting with a permission probe. Give Reviewer a read-only session or candidate packet and return findings to Coordinator. ## Candidate tool mapping Use this checklist to inspect each session; tool names below remain candidates until observed. Use only tools the installed client exposes. Check image viewing and code execution separately from text reading and file writing. | Role | Declared observed tools| Reason and permission scope | Capability without an observed name; section 5.7 route | | --- | --- | --- | --- | | Researcher | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `codebase_search`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Read actual sources/repositories, research all domains, write only its library and handoffs, return the research receipt. | Image/page viewing and measurement runner: Image generation/viewing and Code execution/browser. | | Coordinator | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Read artifacts and dated host evidence, write only owned coordination/capture paths, route handoffs. | Visual inspection and worker creation: Image generation/viewing and Separate workers. | | Designer | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Read brand inputs, top up design research, write owned brand/mockup specifications and handoffs. | Image creation/viewing, contrast calculation, responsive inspection: Image generation/viewing and Code execution/browser. | | Graphics | `view_file`, `list_dir`, `find_by_name`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Verify current tool rights and formats, read supplied assets, write owned asset specifications/manifests and handoffs. | Image generation/viewing and export/dimension inspector: Image generation/viewing and Code execution/browser. | | Builder | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `codebase_search`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Read implementation and stack sources, write owned source/build instructions and handoffs. | Dependency/build/test execution, browser/screenshots, response/performance measurements, deploy: Code execution/browser and Deploy access. | | Optimizer | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `codebase_search`, `search_web`, `read_url_content`, `write_to_file`, `send_message` | Inspect source HTML, top up current SEO/AEO, write owned discovery artifacts and handoffs. | Validators, rendered browser/response checks and authenticated analytics: Code execution/browser. | | Reviewer | `view_file`, `list_dir`, `grep_search`, `find_by_name`, `codebase_search`, `search_web`, `read_url_content`, `send_message` | Read candidate and current attack/standards evidence; return findings for Coordinator capture. No filesystem writes, including findings. | Visual/test/scanner evidence inspection beyond text: Image generation/viewing and Code execution/browser; reproduction goes to a capable runner. | ## Settling checks - Open each complete role file; record exact client version and the path each session read. - Probe excluded Reviewer writes and execution through every inherited route; unchanged bytes are required. - Probe image inspection, generation, browser, build and source retrieval separately. - Verify each manually delivered handoff, concurrency limits and the research gate before Coordinator works. - Preserve source bytes and return a path-by-path receipt. Missing capability produces BLOCKED work. - References: [subagents](https://antigravity.google/docs/subagents), [agent manager](https://antigravity.google/docs/cli/commands/agents). ## Check that it loaded Native loading on this host is awaiting a real-client trial; the check below confirms it on yours. Current status: [compatibility](../docs/COMPATIBILITY.md). Give the host a synthetic brief and ask it to identify the entrypoint and complete asset list. Record client version, date, exact path, tool availability and the observed invocation. Start Researcher, verify saved library read-back, then confirm Coordinator waits for its gate. For TEAM, record seven separate routes and families; an inherited family leaves independent review BLOCKED. Never grant a Reviewer write access just to persist findings. Return text for Coordinator to capture. When a tool is absent, keep current research BLOCKED until opened sources return; save and reattach named file bodies before the save gate; return inspected images before the visual gate. Read the [method](../docs/METHOD.md), [compatibility](../docs/COMPATIBILITY.md) and [installer contract](../docs/INSTALLER.md). --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/grok/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/grok.md Relative links below resolve against that repository source location. --- # Grok TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Set up Bots through the account controls using the complete roles and manual relay below. See [compatibility](../docs/COMPATIBILITY.md) for setup evidence and host trials. - SOLO: paste the starter in [zero-install](zero-install.md), or give a file-capable session the complete canonical skill. - TEAM: install or attach all five [bundles](../docs/bundles/team.md), open the seven sessions yourself, then paste the setup request below in each session. - Native skill (Grok Build CLI): install with `--target grok`, which writes `.grok/skills/website-build-skill/`, then trust the project folder (accept the trust prompt, or run `grok --trust` there once). Grok Build offers project skills only in a trusted folder. Native loading was tested on Grok Build CLI 1.0.40 (macOS, SOLO, trusted folder, 2026-09-24; one run, limitations in the [receipt](../docs/evidence/host-trial-grok-build-2026-09-24.json)); `grok inspect --json` lists the skill with source `project` when it is available to you. Grok Bot is a separate app with its own setup, below. Current status: [compatibility](../docs/COMPATIBILITY.md). - Independence: workers from one model family leave the independent release gate BLOCKED. - Reference: [Bot setup](https://docs.x.ai/grok-bot/bots), [collaboration](https://docs.x.ai/grok-bot/chat-and-collaboration), [skills](https://docs.x.ai/build/features/skills-plugins-marketplaces). Use these references when checking the installed controls. ```text Give me the team (option 2). I open each session myself and carry the handoffs between them, with Researcher first. Use the attached team.md and the complete role body for this session. Return its setup acknowledgement with session reference, model family and tool limits. I collect seven acknowledgements through the manual setup below. Give my request to Researcher first; every other worker waits for the saved-library gate. Use a reviewer from a different model family for independent review. ``` ## Grok Bot Grok Bot is the Bots app (desktop, iPhone and Android). It keeps skills in its own private skill library, so `--target grok` does not reach it. This setup follows xAI's [Grok Bot documentation](https://docs.x.ai/grok-bot/overview), read 2026-09-25; no trial receipt exists yet, so treat each step as the documented route and confirm it on your account. **What the app gives you** - One private skill library shared by all your Bots. Type `/` in the composer to use a saved skill; if one is missing, open **Marketplace → Your plugins → Manage plugins and skills** and look under **Private skills**. - One shared cloud computer for all your Bots, with a browser, a command line and a `/workspace` folder every Bot can read. Keep `site-work/` there. - Bots with their own conversation and memory. A Bot's standing rules live in its profile **description**. - Group chats of two to six Bots, `@` mentions, and direct Bot-to-Bot handoffs. Handoffs posted to a group are text only, so a Bot sends an image directly to the Bot that must inspect it. - Up to six attachments per message, 25 MB per document. - Every Bot runs on Grok: there is no model picker (reported by the operator, 2026-09-25). A reviewer from a different model family therefore sits outside Grok Bot. **SOLO** 1. Choose **New** (`Cmd/Ctrl+N`), then **Create new Bot**, and use **Edit Profile** to name it `Website Builder`. 2. Ask it to install the package into `/workspace` and learn it as a skill: `Run npx website-build-skill@latest --solo --target portable --dir /workspace/website-build-skill --yes, read /workspace/website-build-skill/.website-build-skill/SKILL.md, and save it as a private skill called website-build-skill that reads the files beside it.` The Bot learns the skill and keeps it across conversations (reported by the operator, 2026-09-25). 3. Type `/` and confirm `website-build-skill` is listed. If it is missing, attach the five [bundles](../docs/bundles/team.md) to your first message and ask the Bot to keep them in `/workspace/website-build-skill/`. 4. Start the build with `/website-build-skill` and your site request. **TEAM** - Create six Bots in the order below (Researcher, Coordinator, Designer, Graphics, Builder, Optimizer). Save each role body from the [team bundle](../docs/bundles/team.md) to `/workspace/site-work/roles/.md` and write in each description: `You are Website . Before every task, read /workspace/site-work/roles/.md and follow it.` - Put those six in one group, Researcher first. A Bot keeps its memory, skills and context when it joins a group (reported by the operator, 2026-09-25). Six is the group limit, and it fits: the seventh role, Reviewer, needs a different model family, so it runs outside Grok Bot and you relay the final packet to it. - Follow the steps below for setup acknowledgements, the research gate and the handoff contract. Since every Bot reads `/workspace`, pass file paths in handoffs rather than reattaching files. ## Manual fallback TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. This manual relay is the TEAM workflow. **Prepare the files and workspace** - Get the complete `docs/bundles/team.md`, `method.md`, `prompts.md`, `playbooks.md`, and `checklists.md` from the same release. The team bundle contains the seven complete role bodies and this guide; the prompts bundle contains active prompts. - Get `skills/website-build-skill/archive/graphics-design-original.md` separately for Researcher's provenance reading. Keep its inactive header and do not execute its historical save instructions. - Choose the private project workspace where `site-work/` will live. Keep its research and brand files out of the public skill repository and deployment output. - Open your host's Bots or agents area and create one worker at a time. Give each the entire named role body wherever your host keeps standing instructions (in Grok Bot, a role file in `/workspace` that the description points to, as in [Grok Bot](grok.md#grok-bot)), then save. If your host has no such place, use separate chats and relay the complete role packets manually rather than inventing an API. - Give each Bot the method, prompt, playbook and checklist bundles as files or pasted text through controls your account actually provides. If a filename inside an instruction is inaccessible, attach or paste that file's contents with its label before work starts. Confirm file access separately for each Bot. - Request only a setup acknowledgement: name, received role, route, actual model family if known, tools, and inaccessible files. For all Bots except Researcher add: `Wait for the research gate; do not begin your role yet.` Pass each work gate using its own required evidence. **Create these seven Bots in this exact order** 1. **Website Researcher.** Click **New**, name it `Website Researcher`, paste the full `roles/researcher.md` body from the team bundle, and save. Supply the active website research prompt and inactive graphics original. Check search, URL/repository reads, viewing, and project file persistence. This Bot gets the first work assignment. 2. **Website Coordinator.** Click **New**, name it `Website Coordinator`, paste the full `roles/coordinator.md` body, and save. Give it the other six Bot names and actual routes, the handoff contract below, and the instruction to wait for Researcher's passing receipt. 3. **Website Designer.** Click **New**, name it `Website Designer`, paste `roles/designer.md` in full, and save. Check whether it can receive/view images and produce visual mockups. It waits for research acceptance and the scope-to-designer handoff. 4. **Website Graphics.** Click **New**, name it `Website Graphics`, paste `roles/graphics.md` in full, and save. Record available image/export tools and existing authorized tiers. It waits for research acceptance, then works with Designer on assets and rights. 5. **Website Builder.** Click **New**, name it `Website Builder`, paste `roles/builder.md` in full, and save. Check code/file/browser tools and authorized implementation access. It waits for accepted research and the user's named mockup selection before choosing a stack or coding. 6. **Website Optimizer.** Click **New**, name it `Website Optimizer`, paste `roles/optimizer.md` in full, and save. Check source retrieval, validators and rendered-page access. It works with and after Builder on its bounded discovery assignment. 7. **Website Reviewer.** Click **New**, name it `Website Reviewer`, paste `roles/reviewer.md` in full, and save. Use read-only permissions where the host supports them; otherwise use a read-only artifact packet and returned text. Record the actual model family. If the host cannot supply a family different from Builder's, keep this Bot INTERNAL and relay the final packet to an actual different-family reviewer. The release gate stays BLOCKED until that review happens. **Connect the group and verify its routes** - If your host offers group chats, create `Website Design` and add the saved workers in the same order, Researcher first, up to the host's group limit (six in Grok Bot, where Reviewer runs outside the group). Record the actual route for each. Verify one bounded setup message reaches the intended recipient using the host's documented addressing controls. - Use separate Bot chats and relay each handoff and its attachments manually, including when the account offers a group view. Keep distinct workers on separate routes; separate workers need separate routes, and labels inside one chat stay one worker. Ordinary single-chat work uses SOLO. - If group messages are text-only, attach mockups and assets directly to the receiving Bot. Image review needs the Bot to open the image itself; a text path alone gives it only the path, so confirm the receiving Bot can inspect the image directly. - Give Coordinator this complete contract after replacing each angle-bracket route with the real saved Bot/chat reference. Unresolved placeholders keep setup incomplete. ```text You are Website Coordinator for Website Design. Website Researcher is agent one and owns the shared research library. Do not begin work until its passing research receipt arrives. Your first job then is to read its files and validate that gate. Routes: Website Researcher: Website Designer: Website Graphics: Website Builder: Website Optimizer: Website Reviewer: Your Coordinator route: Project workspace: Record actual routes, families, tools, and ownership after research acceptance. Do not infer family or permissions from a Bot name. Keep one writer per artifact. Researcher alone writes site-work/research/library/. Every other role reads its named library files first, learns the domain checklists, and tops up its own domain. Send stale or wrong claims back to Researcher; pause affected work for correction. The research gate requires saved files for all touched domains and an opened source URL, publication/update date or explicit unknown, and access date on or after the ASK DATE for every claim we will act on. UNVERIFIED placeholders cannot pass. Then confirm my brief and send scope-to-designer and scope-to-graphics. Require my named mockup selection before selected-design-to-builder. Send scope-to-optimizer with Builder's route/content inputs. Send the fixed final candidate to Reviewer only after pre-review gates pass. Different-family review remains mandatory. Read every returned artifact. Preserve Reviewer returns unchanged. Use every field in templates/handoff.yaml and record gate acceptance only in site-work/status.md. Return every complete addressed packet for me to relay manually. ``` **Trigger the first work handoff** Send this directly to **Website Researcher**: ```text Begin user-request-to-researcher. You are the first working agent. My site request: My authorized project workspace: Supplied materials: Use your full role and prompts/01-website-deep-research.md to cover all fourteen research domains current as of the ASK DATE. Ask for only missing inputs needed to research. Write the grouped library, README headlines, per-domain checklists, source/date and disagreement ledgers, surprises, memory proposal/receipt, and staleness schedule. If you cannot search, say so, provide query-plan.md and unresolved-claims.md, and leave the research gate BLOCKED. If you cannot save files, return named file bodies for me to save and reattach; wait for read-back before passing the disk gate. Pass only when the saved library covers every touched domain and every claim we will act on has an opened source URL, a publication/update date or explicit unknown, and an access date on or after the ASK DATE. Read back the actual saved files before confirming persistence. Then return research-to-coordinator addressed to for me to relay, using all handoff fields and readable artifacts. No other role starts before this research gate. ``` - First receipt: Researcher supplies `site-work/handoffs/researcher/research-.yaml` and its library. Copy the whole receipt and attach the library to Coordinator yourself. - First acceptance: Coordinator opens the files, checks coverage and source dates, and writes PASS or BLOCKED in `site-work/status.md`. Only PASS releases downstream scope handoffs. - First design: Designer and Graphics read their named library files, return learning/top-up receipts, and apply their domain checklists. They open the evidence behind the research summary before starting. - Persistence: preserve the files and receipts, and reattach the current revision in each new session. Carry state in the saved files, since a Bot may not remember a group chat. A durable-memory proposal remains pending until an actual supported save is read back. - Completion: report seven saved definitions, seven setup acknowledgements, actual routes and families, research-gate status, and all remaining capability limits separately. Never call the whole team ready from definition creation alone. --- Canonical page: https://website-build-skill.aunysillyme.dev/docs/chatgpt-project/ Source: https://github.com/aunysillyme/website-build-skill/blob/7ba900046cb9f5d8ff0b3077924d35d07cbff122/adapters/chatgpt-project.md Relative links below resolve against that repository source location. --- # ChatGPT Project TEAM: one command installs the seven workers' files; you open each session yourself and carry the handoffs between them. Researcher goes first. Use an eligible account's Project instructions and file controls. Paste the instruction block below into project settings and attach the five bundles. This selects SOLO. Configure the Project through its account controls before starting. ## Instruction block ```text Use option 1, Teach my AI, with the attached website-building method for this project. Read method.md first, then give the request and tool limits to Researcher in team.md. Complete its full sweep and save site-work/research/library/ before any other job. Without file tools, return named bodies for me to save and reattach before the gate. Every later role reads its named library files first and tops up its own domain; flag stale or wrong claims to Researcher and pause affected decisions. Use checklists.md to record evidence for the current stage. Research current claims before relying on them; cite the opened source and date. Extract the user's brand or request their profile-picture and banner screenshots. Show three visual design directions and wait for a named selection before site code. Choose the smallest stack that meets the approved requirements. Load playbooks.md only for the topic being implemented or checked. Keep one primary action and truthful content consistent across every route. Never describe a proposed file, test, memory write, or deployment as completed. For each tool the next stage needs, use it or return the exact handoff for a capable runner. Follow the human's recorded install choice; this upload selects SOLO. A later switch to TEAM installs the seven workers' files in one command; you open each session yourself and carry the handoffs between them, with Researcher first. Read all seven complete role bodies in team.md on either path. Before each job, complete its learning and dated research; save and reopen handoffs, including self-handoffs. Name the previous framing you are clearing when changing solo jobs. An independent pre-release reviewer must use a different model family from the builder. Keep project data in this project; do not upload it to public examples or repositories. Do not publish beyond the user's recorded authorization. Finish each stage with artifacts, completed checks, next steps, and the next owner. ``` ## Upload these files from one revision - [method.md](../docs/bundles/method.md) - [prompts.md](../docs/bundles/prompts.md) - [playbooks.md](../docs/bundles/playbooks.md) - [checklists.md](../docs/bundles/checklists.md) - [team.md](../docs/bundles/team.md) Supply private brief/brand files separately in the user's own project. For provenance only, supply the [inactive archive](../skills/website-build-skill/archive/graphics-design-original.md) separately if needed; never execute it. ## Check that it loaded Native loading on this host is awaiting a real-client trial; the check below confirms it on yours. Current status: [compatibility](../docs/COMPATIBILITY.md). Give the host a synthetic brief and ask it to identify the entrypoint and complete asset list. Record client version, date, exact path, tool availability and the observed invocation. Start Researcher, verify saved library read-back, then confirm Coordinator waits for its gate. For TEAM, record seven separate routes and families; an inherited family leaves independent review BLOCKED. Never grant a Reviewer write access just to persist findings. Return text for Coordinator to capture. When a tool is absent, keep current research BLOCKED until opened sources return; save and reattach named file bodies before the save gate; return inspected images before the visual gate. Read the [method](../docs/METHOD.md), [compatibility](../docs/COMPATIBILITY.md) and [installer contract](../docs/INSTALLER.md).