Skip to main content
Cx
Start building
Skip the story

Releases to Claude Code, OpenClaw and your own code

Your experts write what your AI agents know.

Support leads, ops and legal teams write skills in plain language. Contexory checks every release against your own cases, then delivers it to your agents. If one goes wrong, you roll production back to the last good release.

Start free trial

14 days of Team · no credit card · See pricing

Works with

  • Claude Code

    Plugin on npm

  • OpenClaw

    Plugin, also on ClawHub

  • GitHub

    Skills sync to your repo

  • Your own code

    Signed API and SDK

Scroll, and the agent fills one layer per section

Part 1 · In its head

What it knows.

Four kinds of context fill the agent's head. Your team writes each one in plain language.

01 / 08Instructions

Do I need to know YAML or frontmatter?

Who it is, and how it behaves.

An agent first needs a voice and a job. Instructions say who it serves, what it may promise, and when to hand a customer to a person.

Example
AGENTS.md· Support concierge · instructions
  1. 1You are the support concierge for our store.
  2. 2Answer refund and order questions politely.
  3. 3Always look up the order before promising anything.
  4. 4Escalate to a human when the customer is upset twice in a row.
What Contexory doesYou write in plain language. Contexory lays out the files every host expects, so nobody writes YAML or frontmatter.
Write your first instructions
02 / 08Skills

Can the people who know the work write this?

Playbooks it reaches for when they fit.

Skills hold the know-how your best people carry in their heads. The agent loads one only when a conversation needs it.

Example

  • Refund reply

    Loads when a customer asks for a refund or a return.

  • Ticket triage

    Loads when a new ticket arrives in the queue.

What Contexory doesAn assistant turns a playbook into a skill, a trigger test checks the agent picks it up, and one library lets every agent reuse it.
Start your first skill
03 / 08Knowledge

Where do our policies and guides go?

The reference files its skills read.

Policies, tone guides and regional rules live as files beside the skill. The agent opens one only when a step points to it.

Example
  • refund-reply/
  • SKILL.md
  • references/
  • policy.md
  • tone-guide.md
  • regions/
  • eu.md

Reference files travel with the skill.

What Contexory doesReference files sit in folders next to the skill and ship with it to every host.
See how skills and files fit together
04 / 08Tools

What stops it changing things it should not?

What it may reach, and what it may change.

Useful agents look things up and sometimes act. Each tool says plainly whether it only reads or changes something.

Example
  • Find an orderReads
  • Refund an orderChanges things

Unclassified tools stay off. You connect your own account, nothing is stored.

What Contexory doesEvery tool gets a plain name and a reads or changes label. Anything unclassified stays off, and Contexory stores no plaintext credentials.
How tool access is handled

Part 2 · Around it

What makes it ready.

These four add no context. They make the context it has trustworthy: checks, a sealed release, delivery to each host and feedback from runs.

05 / 08Checks

What if an edit makes the agent worse?

Real cases, before anything goes live.

Checks test what the agent already knows. They run against cases your team wrote, and a failing case holds the release out of production.

Example
  • Refund after 30 daysFailedPasses after a fix
  • Where is my orderPasses
What Contexory doesA failing release stays out of production until it is fixed, or until an admin overrides the gate and writes down why.
Read the docs
06 / 08Release

Can we undo a bad change?

Everything ships together, sealed.

What it knows becomes one numbered release that never changes. Production points at it, like a badge it wears.

Example
  • r2Sealed
  • productionr2Live
  • r1kept, unchanged
What Contexory doesOne release for instructions, skills, knowledge and tools. Rolling back moves production to the last good one.
How releases and labels work
07 / 08Delivery

Does it work with the agent we already run?

Written once, plugged in everywhere.

The same release reaches every place the agent runs, in the layout each host expects.

Example
  • Claude CodePlugin: /plugin install contexory@contexory
  • OpenClawPlugin, also on ClawHub
  • Signed APIYour code fetches the release a label points at
What Contexory doesHost-agnostic by construction: the plugins and the signed API deliver exactly the release you approved.
Install for Claude Code
08 / 08Live

How will we know it worked?

Real use makes it better.

Runs flow back, so you see what the agent did with each release.

Example

Release r2 · production

Each run names its release. Compare two releases on the same kind of run.

What Contexory doesEvery run names the release it used, and you can compare two releases side by side.
Start free trial
Context 100% · Ready 4/4

From empty to useful.

Start free trialTalk to us

14 days of Team · no credit card

What it knows

  • 01InstructionsPlain-language instructions, laid out for every host.
  • 02SkillsKnow-how becomes tested skills in one shared library.
  • 03KnowledgeReference files in folders beside each skill.
  • 04ToolsReads or changes, in plain words. Nothing unclassified is on.

What makes it ready

  • 05ChecksA failing release stays out of production.
  • 06ReleaseOne numbered release that never changes, and a rollback to the last good one.
  • 07DeliveryOne release, delivered to every host.
  • 08LiveEvery run names the release it used, so you can compare two.

Three people decide

The person who writes, the person who ships and the person who pays each get a next step.

For the expert who writes

Write the playbook once, in your own words.

  • An assistant handles structure, frontmatter and file layout
  • Edit with your team, live, with version history
  • Review changes before they publish
Start free trial

For the engineer who ships

Production only moves when you move it.

  • Numbered releases that never change once cut
  • Your eval cases run before production moves
  • Runs report which release they used
$ /plugin marketplace add https://www.contexory.com/claude/marketplace.json
$ /plugin install contexory@contexory
Read the Claude Code guide

For the person who pays

Per-seat pricing you can see today.

  • Team is $15 per person a month
  • Owner, Admin and Member roles
  • Email us for volume seats
See pricing

Check it yourself

Everything here is something you can look at before you sign up.

Try Skill Doctor, no account needed

  • Plugins you can inspect

    Both plugins are published on npm, and OpenClaw's is also on ClawHub. Read what they install before you install them.

    @contexory/claude-code-plugin

  • Plain Markdown you own

    Instructions and skills are ordinary Markdown files. Sync them to your own repo through GitHub or export them whenever you like.

    SKILL.md · AGENTS.md

  • A free check, no account

    Paste a SKILL.md into Skill Doctor and see what an agent will misread. It runs in your browser.

    Skill Doctor

  • Docs written for both readers

    Install guides for each host, and a field-by-field SKILL.md reference, all public.

    Docs · SKILL.md reference

See it in action

Watch a skill get written, reviewed and synced.

Three short walkthroughs, under a minute each. No narration.

AuthorA sentence and a runbook become a drafted SKILL.md, audited before anything is created.

Pricing

Priced per person. Start with a 14-day Team trial.

No credit card up front. You pick a plan when the trial ends.

Individual

$9/ month

For one expert getting started.

  • 1 seat
  • AI-assisted authoring & edit suggestions
  • Version history & change overlay
  • Unlimited personal skills
  • GitHub sync: push skills to your repo
Start free trial

Generous usage built in, then simple pay-as-you-go.

Team

Recommended
$15/ seat / month

For teams shipping together.

  • Unlimited seats
  • Everything in Individual, plus:
  • Real-time collaborative editing & presence
  • Skill reviews & approval workflows
  • Team activity feed
Start free trial

Generous usage built in, then simple pay-as-you-go.

Enterprise

Custom

For organizations with procurement and volume needs.

  • Everything in Team
  • Volume seat discounts
  • Security review and procurement, by email
Talk to us

You only pay for teammates when you add them — seats are billed per person.

See every plan line by line →

Security

What a security reviewer asks first, answered here.

We do not have a SOC 2 report yet. The security page says what we do today, and we will walk your reviewer through it.

Read the security pageEmail us about a security review

  • No stored credentials

    Each person connects their own account through a broker, so Contexory holds no plaintext tool credentials.

  • Releases are an audit trail

    A release never changes once cut, and every label move is recorded.

  • AI providers' terms apply

    Whether OpenAI or Tavily may train on content sent to them is set by each provider's own terms. LangSmith tracing is off unless you turn it on.

  • Roles enforced in the database

    Owner, Admin and Member roles, backed by row-level security.

  • Encrypted in transit and at rest

    HTTPS, with TLS terminated by our hosting provider (Vercel). AES-256 at rest, payments through Stripe Checkout.

Questions

What people ask before they start.

Do I need to be technical?
No. You write in plain language and the assistant handles frontmatter, structure and file layout. An engineer sets up the release and delivery side once.
Which agents does it work with?
Claude Code and OpenClaw through their plugins, any repo through GitHub sync, and your own code through the signed API and SDK.
What happens when my trial ends?
You pick Individual or Team to keep your gated features, or stop. There is no card on file during the trial, so nothing is charged unless you choose a plan. Nothing is deleted; your instructions and skills stay yours.
Can engineers keep control of production?
Yes. Experts edit; production only changes when someone with access moves the production label to a new release. The eval gate holds a failing release out of production unless an admin overrides it with a reason.
Is our content used to train AI models?
That is set by each AI provider's own terms. Assistant requests go to OpenAI and Tavily, and LangSmith tracing is off by default. The privacy policy lists every subprocessor.
Can we leave and take our skills with us?
Yes. Skills and instructions are plain Markdown. Sync them to your own repo through GitHub or export them at any time.

Give your agent what your team knows.

Write the first skill today. Your engineers keep releases, checks and rollback.

Start free trialTalk to us

14 days of Team · no credit card