Trend Detector
Compares current sprint signals to the prior 4–6 sprints. Flags trajectory not just snapshot — three consecutive declines is treated as a leading indicator regardless of absolute value.
When this skill fires
·ARIA generates a sprint report andSprint Context Loader has loaded prior sprints ·An engineer asks:"is this getting better or worse?"
·A regression alert is triggered by a metric crossing a threshold
Why trend > snapshot
→A 91% test pass rate sounds great in isolation. If it was 96% three sprints ago, it's a leading indicator of a problem that hasn't surfaced yet.
Trend detection catches problems while there's still time to fix them. Snapshot reporting catches them when they've already become incidents.
What gets trended
Across all agent metrics:
·Test count (growing / shrinking / stable)
·Pass rate
·Flaky test ratio
·Time-to-green on PRs
·Bend point (concurrency at SLO breach)
·Tail latency (p99, p99.9)
·Soak findings count
·Open Critical / High findings
·Time-to-remediate
·New findings per sprint
·Scenario coverage % of acceptance criteria
·Ambiguity flag count
Cross-agent
·Mean cycle time
·Rework rate (tests added then removed)
Trend classification
For each metric, the detector classifies the trend as:
·Improving — three consecutive sprints in the right direction
·Stable — no significant change
·Degrading — three consecutive sprints in the wrong direction
·Volatile — alternating, no clear direction (often a signal of noise rather than trend)
·New — not enough history (< 3 prior sprints)
Three is the threshold because two-point trends are noise; three-point trends are signal. This rule was set after early ARIA reports flagged trends that reverted next sprint. Confidence
Each trend has a confidence score based on:
·Variance across the historical window
·Number of data points
·Whether the metric is naturally noisy (flaky test counts) or stable (open findings)
Low-confidence trends are reported but not used to drive recommendations.
What it produces
For each metric:
·Trend class (improving / stable / degrading / volatile / new)
·Confidence (high / medium / low)
·A one-sentence summary:"Test pass rate has declined 5pp over the last 3 sprints; the regression centers on the checkout flow."
The summaries are whatRecommendation Drafter uses to propose actions. The classifications are whatLeadership Summary uses to decide what makes the exec brief.
Example
For a metric "open Critical security findings":
·Sprint -3: 1
·Sprint -2: 1
·Sprint -1: 2
·Sprint 0 (current): 3
Classification: Degrading, confidence High, summary"Critical findings have grown from 1 to 3 over the last 4 sprints. The most recent finding (SQL injection in /api/search) was confirmed by SARA last week."