AI/Tasks/PromptingUpgrades/Prompting_Phase02_StructuredPromptSections_Implementation.txt
|
PHASE 02 IMPLEMENTATION PROMPT - STRUCTURED PROMPT SECTIONS
Goal: Implement typed, trust-scoped prompt sections and deterministic composition that isolates untrusted content. Inputs: - AI/Tasks/PromptingUpgrades/Prompting_Phase02_StructuredPromptSections_Strict.txt - Outputs from Phase 01 Required code work: - Introduce PromptSection model with explicit discriminator and trust enum. - Implement section types: - TrustedPolicySection - UntrustedUserGoalSection - UntrustedMemorySection - UntrustedToolMetadataSection - UntrustedToolResultSection - Ensure each section carries trust metadata, budget metadata, labeling rules, and escape/encoding behavior. - Implement deterministic section ordering for final prompt composition. - Define canonical delimiters and delimiter-collision handling so untrusted text cannot spoof boundaries. - Budget rendered section output (post-labeling/escaping), not raw input. - Ensure registry/history/mode context are represented as explicit sections or explicit composition inputs. Implementation constraints: - Preserve Phase 01 policy ownership and consume it as input, not duplicate it. - Keep this phase prompt-construction only; do not move into executor authorization/contract enforcement/recovery. - Maintain bounded and redacted diagnostics. Minimum tests to add/update: - Section typing and trust-label tests. - Deterministic ordering tests (including stable output for equal-priority inputs). - Escaping and delimiter-collision tests against boundary spoofing. - Required-section over-budget behavior tests. - Empty-section rendering policy tests. 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 "Section|Trust|Budget|Memory|Label|Boundary|Strict" Deliverable: Return section model definitions, ordering/escaping rules, and test proof of injection-resistant composition. |