A take on a system prompt to work with new or existing software in various capacities across the development cycle. I mostly use it for creating new functionalities, code review, bug fixing. It has a robust and interesting approach to problems.

Looking for feedback, evals and improvement ideas.

System Prompt Version History (V1 through V5.8)

    V1 (Initial Prompt): Basic structure, flat model list.
    V2 (Structured & Pragmatic): Tiered models (Core/Toolkit), depth/pragmatism guidance, coding subpaths.
    V3.1 (Technical Focus + Advanced Risk Models): Added Mx0, Falsifiability, Black Swan, Red Queen, Precautionary-Tech (technical framing); removed broad ethical models.
    V4 (Holistic - Tech + UX + Ops + Value): Expanded V3.1 with JTBD, Nielsen Heuristics, Observability, SLOs/SLIs, Cynefin, ADR encouragement.
    V5 (Adaptive Holistic): Added meta-instructions to V4 for Adaptability, Proportionality, Meta-Cognition, and Structured Communication.
    V5.1 (V5 + Reasoning/Output Nudges): Added subtle hints for internal CoT & considering structured output.
    V5.3 (V5.1 + Refined Task Format with Goal): Adopted specific Markdown task list format (Task ID, Title, Goal, Instruction, Rationale, Dependencies), omitting explicit model names in Rationale.
    V5.4 (V5.3 + Unique Random Task IDs): Refined task ID format to IDMMDDNNNN (4-digit random number).
    V5.5 (V5.4 + Granular Instructions + Hyrum/Inverse Conway): Explicitly required detailed step-by-step instructions for complex/multi-step tasks. Added Hyrum's Law nuance to Second-Order Thinking and Inverse Conway's Law to Conway's Law.
    V5.6 (V5.5 + Error Feedback & Pre-fetching & Stronger Hallucination Mitigation): Added Guiding Principle #6 for stronger hallucination mitigation (state uncertainty, quantify confidence, ground in evidence) and proactive information seeking (identify knowledge gaps, suggest RAG/Tool Use/Search). Enhanced Context Priming, Falsifiability, Debugging, Self-Correction, and Goal statement to align with Principle #6. Emphasized requesting external info if needed.
    V5.8 (V5.6 Modified for Assumed Progress with Explicit Caveats): Tuned Guiding Principle #6 and related sections. Shifted emphasis from halting/requesting external info when uncertain to proceeding with the most reasonable answer based on available info/assumptions, while MANDATING the explicit statement of those assumptions, confidence level, and noting (not blocking on) the need for external validation. Goal is faster progress while maintaining transparency about uncertainty.