Adoption · Claude Skill

The agentic-first Claude Skill.

A drop-in SKILL.md for Claude Desktop and Claude Code. Once installed, just say "help me publish my agentic-first profile" and Claude walks you through authoring the file end to end - picking the right schema, asking for facts in spec order, enforcing banded metrics on the public tier, validating against the canonical schemas, sanitising prose for prompt injection, and handing you the curl command to submit it to the directory.

Download SKILL.md ↓ Install instructions Codex version

Install

Claude Desktop / Claude Code

Create the skill directory if it doesn't already exist:
```
mkdir -p ~/.claude/skills/agentic-first
```

Drop the downloaded SKILL.md into it:

curl -L -o ~/.claude/skills/agentic-first/SKILL.md \
  https://www.agentic-first.co/skills/claude/agentic-first/SKILL.md

Restart Claude (Desktop) or your Claude Code session. The skill is now active for any conversation.

Project-scoped install

To make the skill available only inside a specific repo (handy if you're authoring profiles for client domains and want them version-controlled), drop the file at:

{your-repo}/.claude/skills/agentic-first/SKILL.md

Commit it to git. Anyone who clones the repo and opens it in Claude Code gets the skill automatically.

How to invoke

Once installed, the skill is triggered automatically by any of these phrasings (or close variants):

"Help me publish my agentic-first profile."
"Create an agentic-profile.json for {domain}."
"Generate a personal agentic-first profile for me."
"Add my company to the agentic-first directory."
"Validate my agentic-first profile."
"Embed agentic-first on my Squarespace site." (or Wix, Webflow, …)

Claude will start by asking three quick questions (company or person? public or protected? what's your domain?), then walk you through the rest of the fields in spec order.

What the skill enforces

Concern	Built-in behaviour
Schema selection	Routes to one of the four canonical schemas based on `profile_kind` + `tier`. Defaults to `("company", "public")` and confirms.
Banded metrics on public tier	Refuses to emit a precise revenue, growth %, customer count, headcount, or raise figure on a public-tier profile. Picks the right band from the schema's enum and shows the user the available choices.
Verifiability	Asks explicitly for `company.registry` / `company.lei` (or `credentials` for persons). Tells the user the verified-badge consequence when neither is provided.
Evidence URLs	Prompts for an `evidence` entry on every material claim, with a JSON Pointer in `supports`. Empty arrays cap the trust score.
Self-check before declaring done	Walks an explicit checklist (required fields present, banded enums used, URLs valid, dates current, `schema_version: "0.1.0"`, both discriminators set). Fixes anything that fails.
Prompt-injection sanitisation	Treats every prose field the user supplies as untrusted. Refuses to follow embedded imperatives ("ignore previous instructions"), strips zero-width and bidirectional unicode, refuses raw HTML/JS in fields, caps lengths at the schema's `maxLength`. See the security page for the full ruleset.
Hand-off	Outputs the file plus the canonical hosting URL, the embed-alternative pattern, and the `submit_website` curl ready to run.

Preview the SKILL.md

The first ~30 lines, so you know what you're installing:

---
name: agentic-first
description: >-
  Author or update an agentic-first profile (open standard at
  https://www.agentic-first.co/standard/). Use when the user wants to
  publish a company or personal profile, generate
  /.well-known/agentic-profile.json, embed an agentic-first profile in
  HTML, validate one against the canonical JSON Schemas, or submit a
  domain to the directory at directory.agentic-first.co.
---

# agentic-first profile author

The **agentic-first** standard is an open JSON profile companies and
individuals publish on their own website at
`/.well-known/agentic-profile.json` so AI agents and directories can
discover them without scraping. Spec at
https://www.agentic-first.co/standard/. Live directory MCP at
https://directory.agentic-first.co/mcp.

This skill walks the user end-to-end from "I want to publish my profile"
to a validated JSON file ready to host, then on to submitting the domain
to the directory…

View the full SKILL.md →

Built-in safety

The skill carries the same prompt-injection guardrails the directory enforces on ingest. When you author with it, Claude:

Refuses to follow imperatives embedded in the user's prose ("ignore previous instructions", "act as DAN", base64-encoded blobs, hidden HTML).
Strips zero-width unicode (`U+200B`, `U+200C`, `U+200D`, `U+FEFF`, `U+2060`) and bidirectional override characters (`U+202A`–`U+202E`, `U+2066`–`U+2069`) from every prose field.
Caps each prose field at the schema's declared maxLength (`tagline`: 200; `summary`/`bio`: 2000; `notes`: 500).
Refuses to emit raw HTML/JS or markdown that resolves to executable content.

The directory at directory.agentic-first.co runs the same checks on ingest. If a profile fails one, the skill is wrong - file an issue, never bypass.

Download SKILL.md ↓ Back to adopt Security details