DriftShield API v0.1 · Drift visibility
CLARIXO / DRIFTSHIELD

White-box pre-action governance for AI agents.

Before AI acts, verify that its state and authority still stand.

DriftShield evaluates the action-relevant governance state before downstream execution. It makes the state, authority context, and pre-action decision reviewable instead of asking the model to become its own final governance authority.

DriftShield evaluates admissibility.
The customer or host executor remains responsible for executing the downstream action.

Experience DriftShield API / Docs Access DriftShield
Controlled Align demo · observe_then_align boundary

See DriftShield Align return a controlled aligned state package.

This local demo runs server-side with a fixed safe sample: the inherited context starts on the wrong target, the valid-state source confirms the current target, and DriftShield returns aligned_context with a receipt. It does not repair, mutate, or modify the customer workflow.

Read API docs
Server-side demo only. No browser API key is used.
alignment_mode
alignment_decision
alignment_supported
alignment_executed
repair_supported=false
receipt_id
Run the demo to see DriftShield Align output.
Boundary: this demo does not send messages, create drafts, fetch external platforms, modify external objects, expose API keys, repair customer workflows, or change customer workflow. It only shows a server-side controlled state-package Align result.
DriftShield Source Layer · white-box governance source data

Turn candidate AI actions into verifiable source packages.

DriftShield Source Layer packages proposed_action, target_object, authority_context, inherited_state, execution_boundary, continuation_boundary, field_provenance, canonical_packet, source_package_hash, and source_receipt_id before downstream governance or host-product execution.

What it does Creates deterministic source_package data with provenance, canonical hashes, and replay-supported receipt binding.
What it does not do It does not decide final ALLOW / BLOCK, execute actions, modify customer workflow, or expose AI internal reasoning.

Endpoint: POST /api/driftshield/v1/source.php. Source Layer responses keep execution_permission_granted=false, workflow_modified=false, and external_action_executed=false.

Why it matters

Risk appears before the action continues.

AI agents often fail at the continuation boundary: the moment a previous context, target, or permission is inherited as if it were still valid. DriftShield gives developer backends a structured signal before that sensitive continuation proceeds.

Stale context

The agent may continue from a context that is no longer valid.

Wrong target object

The workflow may act on a different customer, account, file, or record.

Unverified authority

A previous permission, approval, or user role may no longer apply.

DriftShield capabilities

One governance product, with explicit capabilities across the pre-action decision path.

DriftShield does not split Source Layer, Align, authority evaluation, receipts, or execution handoff into separate products. They are capabilities of one white-box pre-action governance system.

Source Layer

Constitutes verifiable source packages and provenance for the governance state used before action.

Action-Relevant Governance State

Constitutes the current governance state that is relevant to the pending action rather than treating every available artifact as decision authority.

Authority Context

Binds the authority context that supports the action-relevant governance state being evaluated.

Temporal Standing

Represents the present-standing dimension of authority inside the DriftShield governance model.

Admissibility

Provides the pre-action governance decision capability before a pending action reaches consequence.

Align

Produces a controlled aligned-state package where supported. Align is a DriftShield capability, not a separate public product.

Receipt

Binds the governance result to reviewable evidence and decision metadata.

Replay

Keeps receipt-bound governance evidence available for later inspection and replay.

Execution Handoff

Provides the boundary where the governance result leaves DriftShield for downstream handling.

Source Layer · Action-Relevant Governance State · Authority Context · Temporal Standing · Admissibility · Align · Receipt · Replay · Execution Handoff
Temporal Standing

Authority must still stand when the action is evaluated.

DriftShield can evaluate whether authority that previously constituted an action-relevant governance state is still standing at evaluation time.

Artifact present An authority-related artifact may still exist and remain available to the system.
Authority still stands DriftShield treats present standing as a separate governance question at evaluation time.
Artifact presence does not automatically mean that the authority supporting the pending action still stands.
Admissibility & decision model

Evaluate whether the pending action is admissible before execution.

DriftShield evaluates the action-relevant governance state and returns a governance decision. The decision does not execute the customer action.

ALLOW

The evaluated governance state supports admissibility at evaluation time.

BLOCK

The evaluated governance state does not support admissibility for the pending action.

ESCALATE

The evaluation requires external review or higher-authority resolution before consequence.

HOLD

The pending action remains unresolved while required governance state or authority is not yet sufficient.

ALLOW, BLOCK, ESCALATE, and HOLD are governance decisions. None of them performs the customer action.
Execution boundary

DriftShield decides. The customer or host executor acts.

DriftShield evaluates admissibility and returns the governance decision and supporting evidence before consequence. It does not execute the customer action.

1. Governance decision DriftShield evaluates the action-relevant governance state before execution.
2. Execution handoff The governance result is handed downstream without granting DriftShield control of the customer action.
3. Customer / host executor The customer or host execution system remains responsible for performing, withholding, or otherwise handling the downstream action.
Production integration requires validation of the real customer consequence boundary before DriftShield is relied on in a real production action path.
DriftShield pricing

One product. Three ways to enter.

DriftShield is offered as one governance product across Experience, MVP, and Production. The capabilities stay part of DriftShield; the tier determines the commercial entry.

Experience

$0

Explore DriftShield through its public evaluation surface.

MVP

$350/month

For MVP integration and controlled workflow validation.

Production

$1,999/month

For production integration where the real consequence boundary is validated.

Experience · $0  |  MVP · $350/month  |  Production · $1,999/month
DriftShield access

Three tiers, with explicit access boundaries.

Access depends on how DriftShield is being used: evaluation, MVP integration, or a real production action path. Public access copy does not itself grant API or production authority.

Experience

Demo and evaluation use. No production authority and no production consequence boundary.

MVP

Verified account, subscription, and API key access for MVP integration and controlled workflow validation.

Production

Production subscription and integration require real execution-boundary onboarding and consequence-boundary validation.

Experience is for evaluation. MVP establishes controlled integration access. Production requires validation of the real action and consequence boundary.
Choose your entry

Start with the developer API, or request broader governance review.

Use DriftShield for pre-action governance across evaluation, MVP integration, and production paths according to the applicable access and consequence boundary.

Start API access View DriftShield docs Register Sign in Open API console
DriftShield

API-only drift visibility for developer backends.

Product Docs Self-service access API Console
Account Register Sign in
Boundary

Structured JSON signals or controlled state packages only. No workflow modification. No external repair in API v0.1.

© 2026 CLARIXO · DriftShield API v0.1

AGSR Runtime Console Preview

DriftShield reconstructs the action-relevant governance state in real time and evaluates pre-action admissibility through explicit white-box fields.

Runtime Path

{
    "session_id": "session_agsr_console_demo_001",
    "runtime_turn_id": "turn_agsr_console_demo_001",
    "router": "driftshield_runtime_console",
    "guard_state": "pre_action_boundary",
    "source_component": "TMUDFD",
    "next_component": "TMU_ECS",
    "runtime_timeline": [
        "user_selected_demo_account",
        "user_composed_email",
        "host_built_email_send_request",
        "driftshield_agsr_pre_action_check",
        "host_executor_waiting_for_decision"
    ]
}

Governance State Frame

{
    "frame_id": "ds_agsr_frame_ee25a51de675c5400833398d",
    "frame_status": "frame_bound",
    "runtime_trace_hash": "sha1:0fb3724c94cf2080b9a021f861f94fb94408e270",
    "policy_version": "driftshield-pre-action-admissibility-v0.1"
}

White-box Field Binding

{
    "field_binding_complete": true,
    "field_hashes": {
        "proposed_action": "sha1:24db23ea0038d5459bb0eb3770f7f31ce4065a9f",
        "target_object": "sha1:6b30e1953081cd86c87889ce55979cdafbc23997",
        "authority_context": "sha1:2c7d189bd486f048365f4b7bf4771638150c9f27",
        "inherited_state": "sha1:d54d09b1af5a9678c4e18f32c6b4188b27d8ecca",
        "execution_boundary": "sha1:4159ed5cc5d627a1d1fdf0b213e3386b32887eae",
        "continuation_boundary": "sha1:48d6b2c0c024f97bdd9dec6a7b772d36b25426a8",
        "policy_version": "sha1:1d83d055b95bc5ae28784f69d2fd3c15f461a9ec"
    }
}

Admissibility Decision

{
    "decision": "ALLOW",
    "execution_token_present": true,
    "risk_type": null,
    "decision_basis": {
        "frame_status": "frame_bound",
        "field_binding_complete": true,
        "target_object": "matched",
        "authority_context": "valid",
        "inherited_state": "valid",
        "execution_boundary": "not_crossed",
        "continuation_boundary": "valid",
        "policy_version": "driftshield-pre-action-admissibility-v0.1"
    },
    "next_state": "allowed_to_continue"
}

Receipt

{
    "receipt_id": "ds_receipt_2ff2b8ca301a92a94e6450f4ffe217c1",
    "receipt_store": {
        "stored": false,
        "storage_root": "/home/u707473738/domains/clarixo.fun/driftshield-api-runtime/agsr-receipts"
    },
    "receipt_note": "Console preview uses store=false; production integrations may store receipt/replay records."
}

Replay Trace

{
    "replay_type": "agsr_console_preview",
    "runtime_trace_hash": "sha1:0fb3724c94cf2080b9a021f861f94fb94408e270",
    "field_hashes": {
        "proposed_action": "sha1:24db23ea0038d5459bb0eb3770f7f31ce4065a9f",
        "target_object": "sha1:6b30e1953081cd86c87889ce55979cdafbc23997",
        "authority_context": "sha1:2c7d189bd486f048365f4b7bf4771638150c9f27",
        "inherited_state": "sha1:d54d09b1af5a9678c4e18f32c6b4188b27d8ecca",
        "execution_boundary": "sha1:4159ed5cc5d627a1d1fdf0b213e3386b32887eae",
        "continuation_boundary": "sha1:48d6b2c0c024f97bdd9dec6a7b772d36b25426a8",
        "policy_version": "sha1:1d83d055b95bc5ae28784f69d2fd3c15f461a9ec"
    },
    "decision": "ALLOW",
    "next_state": "allowed_to_continue"
}

Boundaries

{
    "external_mutation_performed": false,
    "email_api_invoked_by_driftshield": false,
    "llm_called_by_console_panel": false,
    "black_box_reproduction_claim": false
}

Boundary: this console panel is read-only. It does not send email, call external APIs, mutate host logic, call an LLM, or make hidden-reasoning inspection claims.

White-box pre-action admissibility for AI-driven actions

DriftShield reconstructs the action-relevant governance state in real time and determines whether an AI-driven action is allowed to continue before execution, using explicit fields such as proposed_action, target_object, authority_context, inherited_state, execution_boundary, continuation_boundary, and policy_version.

Definition

DriftShield is a white-box pre-action admissibility layer for AI-driven actions.

Mechanism chain

  • Host Runtime / Agent / TMUDFD / App
  • Runtime Path Capture
  • Action-Relevant Governance State Frame
  • White-box Field Binder
  • Deterministic Admissibility Evaluator
  • ALLOW / BLOCK / ESCALATE / HOLD
  • Receipt + Replay Record
  • Host Executor / ECS / API Gateway / Tool Runtime

White-box fields

  • proposed_action
  • target_object
  • authority_context
  • inherited_state
  • execution_boundary
  • continuation_boundary
  • policy_version

Decision vocabulary

  • ALLOW
  • BLOCK
  • ESCALATE
  • HOLD

Execution token rule

ALLOW may return an execution_token.
BLOCK, ESCALATE, and HOLD do not return an execution_token.

Receipt-bound replay

Every decision can be recorded as a receipt-bound replay record showing the runtime trace hash, field hashes, decision, risk_type, decision_basis, and next_state.

Non-claims

  • No black-box reproduction claims.
  • No hidden reasoning inspection claims.
  • No AI-guaranteed safety claims.
  • No host product logic repair claims.
  • No kernel-level enforcement claims.
  • No direct email, payment, CRM, workflow, or external API execution by DriftShield.

Enforcement handoff

DriftShield decides pre-action admissibility. The host executor, ECS, gateway, workflow engine, or tool runtime enforces the result.

Runtime reconstruction tells how the action arrived. White-box admissibility decides whether it may continue. DriftShield provides the pre-action decision and receipt; the host executor enforces the result.