Security and Privacy

SiteKit is for authorized user automation. It is not a CAPTCHA bypass tool, a credential sharing tool, or a way around access controls.

Automation Boundary

SiteKit automates websites you are authorized to use through the same session a normal browser establishes. It does not defeat access controls.

SiteKit should not:

Solve CAPTCHAs.
Evade bot detection.
Spoof device fingerprints.
Rotate proxies to disguise origin.
Work around rate limits imposed by a site.

This is an operational boundary, not only a policy statement. If onboarding detects an anti-automation challenge, the right result is a clear unsupported site status:

UnsupportedSiteError
  reason: anti_automation_challenge
  message: SiteKit does not proceed past challenge pages or access controls.

That is better than producing a half-working adapter that fails unpredictably in pipelines.

Capture Storage

Raw captures can contain cookies, bearer tokens, PII, regulated records, customer contact information, and sensitive free text.

Default storage:

~/.sitekit/captures/raw/

Raw captures are local, ignored by git, and should be encrypted at rest when possible.

Redacted Fixtures

Generated fixtures are safe enough to commit:

adapters/source_portal/fixtures/redacted/users.json
adapters/source_portal/fixtures/redacted/events.json

Redaction keeps:

Field names.
Types.
Structural shape.
Stable fake IDs.
Representative enum values when approved.

Redaction removes:

Tokens and cookies.
Personal names.
Dates of birth.
Emails and phone numbers.
Confidential free text.
Free text that may identify a person.

Audit Trail

Every generated adapter should record:

Capture IDs used for generation.
Human review answers.
LLM model and prompt bundle used for adapter proposals.
Tests that passed before generation.
Package versions generated from the IR.
Provenance for imported adapters.
Capability declarations for generated transforms and browser scripts.

Writes

Write operations should support preview mode by default:

await destination_portal.records.update({
  recordId: 'rec_4745909',
  location: 'Example Warehouse',
  confirm: false,
});

Approval workflows are built above SiteKit. The adapter's job is to make the payload explicit, deterministic, and testable.

Autonomous Discovery

Autonomous browser exploration needs stricter defaults than human-guided recording:

Restrict allowed hosts.
Set time and action budgets.
Block destructive selectors.
Pause before final submit.
Record screenshots, actions, and model decisions.
Prefer sandbox accounts for write exploration.

The autonomous agent may help discover an adapter. It should not silently commit state-changing operations while onboarding production systems.