AI/Tasks/HeuristicsUpgrades/Heuristics_Phase04_SessionReinforcementLogic_Implementation.txt
|
PHASE 04 IMPLEMENTATION PROMPT - SESSION REINFORCEMENT LOGIC
Goal: Implement bounded, expiring, session-local reinforcement that improves continuity without overriding policy or safety. Inputs: - AI/Tasks/HeuristicsUpgrades/Heuristics_Phase04_SessionReinforcementLogic_Strict.txt - Outputs from Phases 01-03 Required code work: - Implement session records for last success and last failure with explicit task/context keys. - Implement continuity boosts and avoidance penalties with hard maximum contribution limits. - Implement expiry and reset rules (time window, turn count, context change, safety-context change). - Implement failure-loop detection and conservative de-escalation path. - Apply reinforcement only as marginal advisory input after cost adjustment and before tie-break. Implementation constraints: - Reinforcement is session-local and advisory only. - Reinforcement must not alter authorization/refusal/confirmation/sandboxing outcomes. - Redact sensitive data in any reinforcement diagnostics. Minimum tests to add/update: - Expiry/reset behavior tests. - Continuity and avoidance bound tests. - Failure-loop suppression tests. - Tests proving reinforcement cannot override safety gates. Validation commands: - dotnet build src/TechToolbox.Agent/TechToolbox.Agent.csproj -c Release - dotnet test src/TechToolbox.Agent/Tests/TechToolbox.Agent.Tests.csproj -c Release --filter "Session|Reinforcement|Stickiness|Continuity|Avoidance" Deliverable: Return record shape, expiration policy, and test evidence for bounded advisory behavior. |