load-anomaly-guard

name	load-anomaly-guard
description	Detects unsafe training load spikes (>20-30% week-over-week) and emits safety flags. Use in nightly background jobs or when reviewing weekly training volume with conservative adjustment recommendations.
metadata	{"short-description":"Background load monitor that flags spikes and proposes protective changes.","agent":"cursor"}

When Cursor should use this skill

• Nightly background check on training data
• Immediately after a high-intensity or long run is logged
• When analyzing weekly training load patterns for safety issues
• When implementing load monitoring features or debugging training load calculations

Invocation guidance

1. Provide recent TrainingHistory, planned Plan window, and any injury flags.
2. Compute week-over-week changes and monotony; flag spikes > deterministic caps.
3. Calculate Acute:Chronic Workload Ratio (ACWR) and training monotony.
4. Suggest adjustments (rest/swaps) and emit SafetyFlag[].
5. Use deterministic thresholds from v0/lib/plan-complexity-engine.ts.

Input schema

See references/input-schema.json.

Output schema

See references/output-schema.json.

Load metrics

Weekly Volume

• Calculation: Sum of all running duration/distance in past 7 days
• Safe increase: <20% week-over-week
• Caution: 20-30% increase
• Danger: >30% increase

Acute:Chronic Workload Ratio (ACWR)

• Acute load: Average daily load over past 7 days
• Chronic load: Average daily load over past 28 days (rolling 4-week average)
• ACWR: Acute / Chronic
• Safe zone: 0.8-1.3
• Caution zone: 1.3-1.5 or 0.5-0.8
• Danger zone: >1.5 or <0.5

Training Monotony

• Calculation: Mean daily load / Standard deviation of daily load
• Low: <1.5 (good variety)
• Moderate: 1.5-2.0 (acceptable)
• High: >2.0 (concerning - too repetitive)

Training Strain

• Calculation: Weekly load × Monotony
• Low: <1000
• Moderate: 1000-2000
• High: >2000 (injury risk)

Detection rules

Spike detection

1. Calculate current week volume
2. Calculate previous week volume
3. If increase >20%, emit load_spike warning
4. If increase >30%, emit load_spike danger

Monotony detection

1. Calculate daily load variance over 2 weeks
2. If monotony >2.0, emit high_monotony warning
3. Suggest adding variety or rest days

ACWR violation

1. Calculate 7-day and 28-day average loads
2. If ACWR >1.5, emit acwr_high danger
3. If ACWR <0.5 (after training), emit detraining warning

Consecutive hard days

1. Count hard sessions (HR >80% max or RPE >7) in past 7 days
2. If 3+ hard days without 48h recovery, emit inadequate_recovery

Integration points

• Background job: Nightly cron (via v0/lib/backgroundJobs.ts if added)
• API: v0/app/api/plan/load-guard (to add) returning flags + suggested adjustments
• UI:
  • Badge on Plan/Today screens showing load status
  • Push/email via v0/lib/email.ts
  • Warning modal before high-load workouts
• Database:
  • Calculate from runs and workouts tables
  • Store load metrics in training_load table (if added)

Safety & guardrails

• If spike >20–30% week-over-week, emit load_spike and recommend rest or reduced volume.
• If injury signals present, bias toward rest-day adjustments.
• No medical diagnosis; advise professional consult on repeated spikes or pain.
• Always prefer conservative adjustments.
• Never suggest increasing load to "make up" for previous low weeks.

Adjustment recommendations

For load spike (20-30%)

• Reduce next week's volume by 15-20%
• Add extra rest day
• Convert 1 hard session to easy
• Maintain intensity but reduce duration

For load spike (>30%)

• Mandatory rest day insertion
• Reduce volume by 30-40% next week
• All runs at easy pace
• Consider cross-training substitution

For high monotony (>2.0)

• Add variety: mix distances, paces, surfaces
• Include cross-training day
• Vary workout types more
• Add optional rest day

For ACWR >1.5

• Immediate volume reduction
• Extra recovery day
• Monitor for injury signals
• Extend build phase timeline

For inadequate recovery

• Insert rest day before next hard session
• Convert upcoming hard session to easy
• Increase recovery time between efforts
• Monitor sleep and soreness

Telemetry

• Emit ai_skill_invoked, ai_safety_flag_raised, and optionally ai_adjustment_applied when suggestions are auto-applied with:
  • load_spike_percent
  • acwr_value
  • monotony_value
  • safety_flags (array)
  • adjustment_recommended
  • auto_applied (boolean)

Common edge cases

• First week of training: No comparison data, establish baseline
• Return from break: Low ACWR is expected, don't flag as detraining
• Taper period: Low load is intentional, don't flag
• Race week: Allow slight spike for race day
• Illness/injury: Pause monitoring during recovery
• Data gaps: Don't flag if data is incomplete

Testing considerations

• Test spike detection with various volume increases
• Verify ACWR calculation accuracy
• Test monotony calculation with different training patterns
• Validate adjustment recommendation appropriateness
• Test with missing or incomplete data
• Verify no false positives during taper or planned down-weeks