JinDaGe - plan-adjuster Skill Details

When Cursor should use this skill

Nightly job or immediately after a run is logged
When the user reports fatigue/injury or requests easier/harder weeks
When performance data indicates plan adjustment is needed
When implementing adaptive training features or debugging adjustment logic

Invocation guidance

Load Plan, Workout, TrainingHistory, and RecentRunTelemetry[].
Apply deterministic ceilings from v0/lib/planAdaptationEngine.ts and v0/lib/plan-complexity-engine.ts before calling the model.
Return Adjustment[], optional RecoveryRecommendation, and confidence.
Only adjust future workouts - never modify completed runs.
Maintain weekly volume within safe limits (±20-30%).

Input schema (JSON)

{
  "profile": UserProfile,
  "currentPlan": Plan,
  "trainingHistory": TrainingHistory,
  "feedback": { "rpeTrend"?: number, "soreness"?: string, "sleepQuality"?: string }
}

Output schema (JSON)

{
  "appliedAt": string,
  "updates": Adjustment[],
  "recovery"?: RecoveryRecommendation,
  "confidence": "low" | "medium" | "high",
  "safetyFlags"?: SafetyFlag[]
}

Integration points

API: v0/app/api/plan/adjust (to add), or chat-triggered adjustments
Logic:
- v0/lib/planAdjustmentService.ts - Adjustment orchestration
- v0/lib/planAdaptationEngine.ts - Adaptive algorithms
- v0/lib/plan-complexity-engine.ts - Safety caps
UI: Plan/Today screens (badge adjusted sessions)
Notifications: v0/lib/email.ts - Email user about significant adjustments
Database: Update workouts table, log adjustments in plan_adjustments (if added)

Safety & guardrails

Never rewrite completed history; adjust only future sessions.
If fatigue/injury signals present, lower intensity/volume and consider rest-day insertion.
Emit SafetyFlag on unsafe load proposals; clamp to deterministic caps.
Maintain at least one rest day per week.
If multiple negative signals, reduce load by at least one level.
Hard stop on pain/injury mentions - recommend rest and professional consultation.

Adjustment types and triggers

Intensity adjustments

Trigger: High RPE trend (>7 for easy runs), poor sleep, soreness
Action: Reduce pace target by 15-30 seconds/km, lower HR zone
Example: Tempo → Easy, Intervals → Tempo

Volume adjustments

Trigger: Missed runs, fatigue, low consistency
Action: Reduce duration by 10-30%, maintain intensity
Example: 60min easy → 45min easy

Session swaps

Trigger: Schedule conflicts, weather, preferences
Action: Reschedule within same week, maintain weekly pattern
Example: Tuesday intervals ↔ Thursday tempo

Rest day insertion

Trigger: Multiple fatigue signals, injury risk, poor recovery
Action: Replace easy run with rest or cross-training
Example: Easy run → Rest day

Cross-training substitution

Trigger: Soreness, minor injury, surface limitations
Action: Replace easy run with cycling/swimming/walking
Example: Easy run → 45min cycling

Telemetry

Emit ai_skill_invoked and ai_adjustment_applied with:
- adjustments_count
- confidence
- safety_flags
- adjustment_types (array of change types)
- user_id (hashed)
- latency_ms

Common edge cases

No adjustment needed: Return empty adjustments array, confirm plan is on track
Conflicting signals: Prioritize safety, default to more conservative option
Major deviation: Consider full plan regeneration instead of adjustments
Peak training: Allow slightly higher load if race is imminent and recovery is adequate
Taper period: Protect taper, only minor adjustments allowed

Testing considerations

Test with various feedback combinations (RPE, soreness, sleep)
Verify load caps are enforced
Test that completed runs are never modified
Validate adjustment rationale clarity
Test with missing data (partial feedback)
Verify SafetyFlag emission for risky adjustments