Skills
@viewfieldai/skills is a free npm package of content-creation skills for coding agents. One command installs slash-command skills for writing landing copy, developing positioning, auditing your site's message structure, and keeping every word free of AI slop, into Claude Code, Codex, Cursor, or any agent that reads the SKILL.md standard.
npx @viewfieldai/skills
Pick the skills you want, pick the agents you use, confirm. No account, no telemetry, no network calls at install time: the skills ship inside the package and are copied into your project as plain Markdown.
The skills are the portable, standalone version of what Viewfield does at full strength, where the same discipline runs on every draft, tuned to your saved brand voice, enforced by the checks on every draft, and informed by everything you've already written.
The five skills
Each has its own page with a full description and a standalone download:
| Skill | Invoke | What it does |
|---|---|---|
| anti-slop | automatic | Writing that reads like a person wrote it: bans the AI tells, enforces specific on-voice prose, self-reviews before returning. |
| marketing-content | automatic | The shared vocabulary (hook, promise, proof, objection, CTA, message home) and channel best practices the other skills build on. |
| grill-me-content | /grill-me-content | A relentless one-question-at-a-time interview about your reader, promise, and proof, then it writes the copy from your answers. |
| repo-to-landing | /repo-to-landing | Turns a technical repo into landing-page copy where every claim carries a citation to the code that proves it. |
| content-architecture | /content-architecture | Audits every prose surface for drift, redundancy, and gaps; presents a visual HTML report; restructures the finding you pick. |
Where things install
| Agent | Destination | Invoke |
|---|---|---|
| Claude Code (project) | .claude/skills/<name>/ | /<name> |
| Claude Code (global) | ~/.claude/skills/<name>/ | /<name> |
| Codex (project) | .codex/skills/<name>/ | $<name> or automatic |
| Codex (global) | ~/.codex/skills/<name>/ | $<name> or automatic |
| Cursor | .cursor/commands/<name>.md | /<name> |
| Agent-Skills standard | ~/.agents/skills/<name>/ | harness-dependent |
Codex reads the same SKILL.md folders as Claude Code. Cursor gets a flattened copy: the skill plus its reference files in one Markdown command file. An opt-in AGENTS.md snippet points agents without a skill system at the installed files.
Project installs anchor to the git root no matter which subdirectory the command runs from: agents discover skills from the repo root, so that's where they belong. Pass --here to target the literal current directory instead.
Commands
npx @viewfieldai/skills # interactive install
npx @viewfieldai/skills add anti-slop --agents claude,codex
npx @viewfieldai/skills list # what ships vs what's installed
npx @viewfieldai/skills update # re-copy installs at this version
npx @viewfieldai/skills remove # uninstall (keeps files you modified)
Installs carry receipts, so the CLI can tell your edits from its own: files you've modified are kept on update and remove unless you pass --force. --dry-run prints the plan and writes nothing.
--agents takes any of claude, claude-global, codex, codex-global, cursor, agents-dir and agents-md; --global is shorthand for the global targets, and --yes skips the prompts (for CI).
The durable artifacts
grill-me-content writes two files at your repo root that every skill (and every future session) reads:
CONTENT.md: reader, promise, proof inventory, objections and their answers.VOICE.md: register, lexicon (we say / we never say), like-this / not-like-this pairs.
When the skills stop being enough
They read VOICE.md and CONTENT.md; Brand holds the same knowledge as structured context with memory behind it, and the checks enforce deterministically what a skill can only ask the model to self-review. When you're ready, connect the MCP and run the same workflows through your brand.