CAPABILITY · CORPUS

A skill compiled once can be safely reused across tenants

Common flows — log in, search, paginate, fill a form — get compiled once and shared through a sanitized corpus, so your agents start ahead.

x-twin-skill-source: corpus
Built for the cost wedge

What cross-tenant skill corpus does

Most browser tasks are not unique. A login on a popular SaaS, a checkout on a major store, a search on a public site — thousands of teams automate the same surfaces. Twin compiles those flows once and, after stripping anything tenant-specific, makes them reusable across tenants. The savings compound across the whole network, not just your own history.

Sanitized by construction

Corpus skills carry only the action structure. Credentials, inputs, and tenant data never enter the shared corpus.

Warm start on common surfaces

Your first run against a widely-automated site can dispatch from a corpus skill instead of paying a cold compile.

Network effect on cost

Every compile improves the corpus, so the marginal cost of automating popular surfaces keeps falling for everyone.

Opt-in and disclosed

Corpus reuse is explicit. You choose whether a tenant draws from and contributes to the shared corpus, and responses disclose the source.

How it works

From a goal to deterministic action

  1. 1Compile locallyA successful run compiles into a tenant skill first, scoped to your workspace.
  2. 2SanitizeA sanitizer strips inputs, secrets, and identifiers, leaving only the reusable action structure.
  3. 3Promote to the corpusGeneric, broadly-useful flows are promoted into the shared corpus with the source disclosed.
  4. 4Dispatch across tenantsOther tenants matching the same goal can dispatch from the corpus skill — a warm start instead of a cold compile.
In practice

See it on a real call

A first-time goal on a widely-automated surface dispatches from the corpus — no cold compile.

run.tsts
const res = await twin.agents.run({
  goal: "Log in and export the last 30 days",
  url: "https://dashboard.popular-saas.com",
  corpus: true, // draw from the shared corpus
});

// → x-twin-cache: hit
// → x-twin-skill-source: corpus
// → x-twin-llm-calls: 0   (warm start)
api.twin-browser.com
  1. Compile locallydone
  2. Sanitizerunning
  3. Promote to the corpusqueued
  4. Dispatch across tenantsqueued
At a glance

What cross-tenant skill corpus is

The facts — how it works, what it costs, and the signal you get back on every call.

PropertyTwin Browser
Shared contentAction structure only
Never sharedSecrets, inputs, tenant data
ParticipationOpt-in per tenant
Disclosurex-twin-skill-source header
IsolationDefault-deny RLS per tenant
BenefitWarm start on common flows
FAQ

Cross-tenant skill corpus — common questions

Could my data leak into another tenant through the corpus?
No. Only the sanitized action structure is shared. Credentials, inputs, and identifiers are stripped before a skill is ever promoted, and tenant data is isolated by default-deny RLS.
Do I have to participate?
No. The corpus is opt-in. You decide whether a tenant draws from it and whether its generic skills are promoted into it.
How do I know when a corpus skill was used?
Every response that dispatches from the shared corpus sets x-twin-skill-source: corpus, so reuse is always disclosed.

Make every run cheaper than the last.

Start free, compile your first skill, and watch the marginal cost per run trend toward zero as your agents reuse what they have already learned.