Notifications and Reporting

The FDE autonomously reports work results via Slack. PMs do not need to ask “report please.” Task completions, anomaly detections, and approval requests are automatically posted to the appropriate channels.

The FDE Reports Proactively

Conventional AI tools follow a passive “answer when asked” model. Deprex’s FDE autonomously performs:

• Task completion notifications: Posts results of scheduled and requested tasks to Slack
• Anomaly detection: Detects and reports KPI fluctuations, errors, and threshold breaches
• Approval requests: Asks PMs to approve automation proposals and skill generation
• Learning reports: Reports new memory additions and detected business patterns
• Error reports: Automatically notifies the dev channel about execution issues

Channel Design

The FDE automatically routes notifications based on content type. The 0ai engineer configures two channels during initial setup.

Channel	Purpose	Content
default_notify_channel	Business notifications	Reports, completion notices, approval requests
dev_channel	Development and testing	Errors, test results, config change notices

.deprex/harness/config/repo-registry.yaml

repos:
  acme:
    dev_channel: "#acme-fde-dev"
    default_notify_channel: "#acme-fde"

If PMs want to change the channel design, they request it from the 0ai engineer.

Notification Types and Automatic Routing

The FDE auto-determines notification type and routes to the appropriate channel.

Notification Type	Content	Auto-Routed To
Task Complete	Scheduled task result report	notify channel
HITL Request	Approval request requiring PM judgment	notify channel
Skill Proposal	Approval request for auto-generated skill	notify channel
Failure	Execution error report	dev channel
Status Update	Progress report for long-running tasks	dev channel
Memory Update	Memory addition/update report	dev channel

HITL (Human-in-the-Loop) Requests are notifications where the FDE asks the PM for judgment on matters it should not decide autonomously.

Notification Format

The FDE generates notifications with a conclusion, evidence, and next action structure.

Task Completion Example

*Weekly KPI Summary - Acme*

Summary:
- Activation rate up +2.1pt week-over-week, trending well
- Support backlog at 48, below alert threshold
- Billing-related inquiries up +30% week-over-week

Needs Review:
- Root cause of billing inquiry increase unidentified. Recommend reviewing #support thread

Source: scheduled task weekly_kpi_summary
Memory: kpi-definitions.md, reporting-rules.md

Approval Request Example

[Skill Approval Request] daily-support-triage

The FDE detected the following task pattern:
- PM manually classifies support tickets every morning
- Classification criteria are fixed (3 categories)
- Results are posted to #support

A skill has been generated. Transition to automated execution?
Shadow Comparison: 95% match with manual classification over the past 5 days

-> Approve / Reject / Observe longer

Error Notification Example

[error] daily_kpi_summary
Situation: Connection to KPI data source timed out
Impact: This morning's KPI summary is incomplete
Action: Auto-retry scheduled in 30 minutes

PM Actions

PMs take three types of actions on notifications:

1. Just Confirm

For task completion and status update notifications, simply reviewing the content is sufficient. No action needed if everything looks correct.

2. Approve or Reject

For skill proposals and automation proposals, respond via Slack.

@fde Approved. But also send a separate notification if backlog exceeds 50 items

The FDE auto-reflects feedback into the skill’s Rules.

3. Provide Feedback

When output quality could be improved, point it out in Slack.

@fde Add category-level breakdown to the KPI summary

The FDE learns from the feedback and reflects it in subsequent skill executions.

owner_user_ids Role

PMs configured in owner_user_ids are the recipients for notifications requiring important decisions (HITL Requests).

owner_user_ids:
  - "U0123456789"  # PM
  - "U0987654321"  # Engineering lead

Limiting to 1-3 decision-makers is recommended.

Important Notes

• Notifications never contain credentials or authentication information
• Error and development notifications are isolated in the dev channel, so business users are not bothered by noise
• To change notification settings, request from the 0ai engineer
• The FDE also autonomously adjusts notification frequency. Repeated identical notifications are aggregated