Configurable email alerts and scheduled digests for enterprise signals and anomalies

Organization administrators and engineering leads can now configure automated email notifications for deterministic anomaly findings and workflow efficiency signals. Instead of manually polling the web dashboard to discover un-redacted secret exposure or runaway agent loops, teams can receive immediate notifications for critical incidents alongside scheduled daily and weekly digests for operational rollups.

Alerts are delivered directly to configured recipient mailboxes or dedicated organization security channels, with authenticated deep links back to the Signals dashboard and session inspection views.

Three delivery tiers

Alert delivery is organized into three distinct cadences based on incident urgency:

1. Immediate alerts (Real-time P0): Triggered when high-risk security findings or destructive agent loops occur. This includes AD-004 (Sensitive File Access) when an agent reads, writes, or scripts against environment files or credentials without data loss prevention (DLP) coverage, and AD-003 (Runaway Session) when an agent repeatedly executes failing commands against policy blocks. To prevent email storms when an agent rapidly loops, immediate notifications are throttled by recipient and finding key with a four-hour suppression window for repeat occurrences unless severity escalates.
2. Daily morning digest (08:00 local time): A consolidated summary covering the previous 24 hours of activity. It surfaces audit-level security notices—such as sensitive file touches that were successfully scrubbed by hook DLP—as well as new proxy operations (AD-006) and non-critical runaway loop warnings. If zero anomalies occurred, it provides an optional confirmation that a clean security baseline was maintained.
3. Weekly executive digest (Mondays): A cross-organization review summarizing workflow efficiency patterns (P1 through P10). It highlights top recoverable token waste—such as session restart churn (P1), model-task mismatches (P10), and context bloat (P5)—along with repeatedly re-read Groundhog files (P2) and overall DLP interception rates.

Zero-exfiltration metadata boundary

Because notification emails travel over external mail transport, payloads adhere to a strict privacy boundary:

• Strictly excluded: Email bodies never include raw secret values, environment variable names, full command outputs, user prompt text, or proprietary source code.
• Included metadata: Notifications convey only allowlisted, structured context: detector key, severity badge, affected session counts, masked pattern categories (such as .env or credentials*.json), access surface (Direct Tool, Shell Command, or Script Mention), and DLP protection state.

Every notification includes direct action buttons linking securely into the Enterprise Signals tab with pre-applied finding filters, allowing engineers to verify the underlying sessions and acknowledge, resolve, or dismiss findings within the platform boundary.

Granular filtering and noise controls

To prevent alert fatigue, both individual members and organization administrators can customize their alert criteria:

• Severity threshold: Recipients can set a minimum severity gate—such as restricting immediate alerts strictly to critical un-redacted findings while ignoring routine warning audits.
• Detector toggles: Specific detectors (such as volume baseline shifts or technology novelty) can be enabled or disabled independently.
• Scope filtering: Team leads can restrict alerts to their managed teams, while individual developers can opt into notifications strictly for their own sessions.
• Project exclusions: Teams working on sandbox repositories, mock environments, or documentation sites can exclude those projects from triggering email alerts.
• Dedicated security routing: Organization administrators can configure a designated security destination (such as an internal security operations inbox) that receives critical alerts unconditionally, independent of personal member settings.

Background and motivation

Previously, anomaly evaluations and signal patterns were visible only when users navigated to the Signals tab in the enterprise portal. For general workflow coaching and token savings analysis, periodic dashboard review is sufficient. However, for acute security events—such as an autonomous agent generating and executing a temporary Python script that loads production credentials—a dashboard-only model created an unacceptable latency window between access and remediation.

Conversely, naively broadcasting an email on every finding creates alert storms that lead engineers to mute or filter notifications entirely. Structuring alerts into immediate critical notices and scheduled daily/weekly rollups ensures urgent risks are seen immediately while routine operational signals are reviewed during standard planning cycles.

Current limitations and next steps

This release focuses on email delivery via standard enterprise mail channels. Organizations requiring real-time streaming directly into SIEM infrastructure can continue to use the dedicated SIEM export pipeline. Native chat webhooks (Slack and Microsoft Teams) and incident management integrations (PagerDuty) are planned for subsequent iterations.

If you have feedback on notification formatting, throttling thresholds, or edge cases in your development workflows, please let us know through the feedback tab.