Skip to content

Remediation Center

3. Remediation Center

The Remediation Center is the automated response and corrective action hub of Theseus Guard. It manages system-generated remediation recommendations — structured proposals by the AI governance engine for corrective actions to address detected drift, compliance gaps, or operational anomalies.

3.1 Overview

The Remediation Center has two primary views:

  • Queue — The main remediation recommendation list and action panel
  • Action Requests — Gray-area authorization requests from agents that need operator sign-off

3.2 Autonomy Policy Widget

A summary widget displays the current Autonomy Policy level for the tenant:

Level Behavior
Manual All recommendations require human review and explicit approval before execution
Supervised Low-risk recommendations within approved confidence thresholds may be automatically executed

The current level, last updated timestamp, any active override end date, and override reason are all shown.

Override Autonomy — Operators can temporarily change the autonomy level with a time-bounded override (e.g., switch to full manual review during an incident). The override requires a reason and an end date.

3.3 Remediation Queue

The queue table lists all pending remediation recommendations:

Column Description
Recommendation ID Unique identifier
Type The category of recommendation (e.g., guard_config_update, agent_suspension, threshold_adjustment, provider_model_auto_set)
Target The guard or agent scope this recommendation applies to
Status pending_review, approved, executing, completed, verified, or timed_out
SLA Time Remaining time before the recommendation must be acted upon
Date When the recommendation was generated

3.4 Recommendation Detail (Side Sheet)

Clicking a recommendation opens a detailed view organized across three tabs:

Overview Tab

  • Severity badge and recommendation type label
  • Recommended action description and rationale summary
  • Confidence score and risk band
  • State Machine Visualizer — A visual flow diagram showing the current state of the recommendation within the remediation workflow (pending_review → approved → authorized → executing → completed → verified)
  • Root Cause Indicator — If the recommendation was triggered by a specific failure condition, the root cause state, reason, and diagnostic details are surfaced here
  • Execution Readiness — A checklist confirming prerequisites are met before the recommendation can be authorized
  • Window Metrics — 7-day and 30-day behavioral metrics for the target scope (event counts, average L1 score, review rate, safe degrade rate, drift delta from baseline)

Analysis & Evidence Tab

  • Guard context panel — current guard configuration and relevant policy details
  • Agent context panel — the agent profile and current operational parameters
  • Agent DNA context panel — the agent's DNA configuration details
  • Correction attempt history — previous automated corrections attempted and their outcomes

Execution Logs Tab

  • A chronological history of all actions taken on this recommendation
  • Each log entry shows the action type, actor, timestamp, and outcome

3.5 Remediation Actions

Depending on the recommendation type and current state, the following actions may be available:

Action Description
Authorize Authorize the system to begin executing the remediation
Start Execution Manually trigger the execution phase
Complete Mark the execution as complete
Verify Confirm the remediation had the intended effect
Rollback Revert the remediation if it caused unintended side effects
Approve Approve the recommendation for processing
Reject Decline the recommendation
Defer Postpone the recommendation
Escalate Escalate to a higher authority
Relax / Tighten Limitations Adjust guard thresholds as part of remediation
Switch Guard Reassign the agent to a different guard
Suspend / Resume Agent Temporarily suspend an agent during remediation
Update Guard Config Apply a targeted guard configuration change

3.6 Workflow Settings

The Workflow Settings modal (accessible from the page header) allows administrators to configure:

  • SLA durations by recommendation type and severity
  • Auto-escalation rules
  • Notification recipients for different recommendation categories

3.7 Action Requests

The Action Requests tab shows gray-area requests submitted by agents operating in autonomous mode that require operator authorization before proceeding.

Each request includes: - Agent name and ID - Action type requested - Evaluation reason (why the agent flagged this as requiring human sign-off) - Request payload - Expiry time - Current status (pending, approved, rejected, expired)

Operators can Approve or Reject each request with a written rationale. Expired requests are automatically closed.