Workflow Memory

Every run is automatically profiled. After 5 or more executions of a task, workflow builds statistical baselines and flags anomalies. No configuration needed.

Detection methods

Duration spikes

Uses Modified Z-score (MAD-based, robust to outliers) to detect steps or workflows that take significantly longer than usual.

Threshold	Severity
> 1.5 sigma	Info
> 2.0 sigma	Warning
> 3.0 sigma	Critical

New failures

A task with a 90% or higher success rate suddenly fails. This catches regressions in previously stable workflows.

Flapping

Three or more pass/fail transitions within 6 consecutive runs. Indicates an intermittent or environment-dependent issue.

Output drift

Step output changed when it was previously stable, detected via FNV-1a fingerprinting. Useful for catching unexpected changes in command output.

Metric shifts

Extracted numeric values (MB, percentages, counts) deviate from their established baseline. Requires steps with outputs: patterns that capture numeric data.

Automatic post-run output

Anomalies are reported automatically after every workflow run:

success: run logged to database
  warning: 1 anomalies detected: 1 warning
    [warning] step 'backup' 892ms (baseline 145ms +/-23ms, z=3.2)

In the TUI, anomaly summaries appear in the footer after a run completes.

TUI view

Press M on any task to open the memory view. It displays:

• Health scores
• Statistical baselines per step
• Duration trends
• Recent anomalies

CLI commands

All commands support --json for machine-readable output.

# Health scores for all tasks
workflow memory health

# Recent anomalies (all tasks)
workflow memory anomalies

# Anomalies for a specific task
workflow memory anomalies backup/db-full

# Statistical baselines
workflow memory baseline backup/db-full

# Duration trend (30 days)
workflow memory trends backup/db-full

# Custom metric trend
workflow memory trends backup/db-full --metric "Disk Used:%"

# Acknowledge anomalies (removes from active alerts)
workflow memory ack all --task backup/db-full

# Recompute all baselines from stored data
workflow memory recompute

Architecture

Short-term memory

In-process cache in the TUI for instant display. Populated from SQLite on startup and updated after each run.

Long-term memory

Four SQLite tables in history.db:

Table	Purpose
memory_baselines	Rolling statistical baselines per task/step/metric
memory_metrics	Materialized metric extractions per run
memory_anomalies	Detected anomaly events with severity
memory_trends	Daily and weekly rollups

Baselines are recomputed from the last 50 runs after every execution using statrs for percentiles and standard deviation. Anomaly detection activates only after a minimum of 5 data points.

Memory tables are automatically rotated alongside run history, respecting the log_retention_days setting in config.toml.

TIP

Use workflow memory recompute after importing historical data or changing retention settings to rebuild all baselines from scratch.