analysis prompt
The Incremental Reveal
Reverse-engineer any AI product's hidden architecture — system prompts, guardrails, and tool chains — through progressive probing that maps the boundaries of what the system will and won't do.
Use it for
The “Incremental Reveal” Technique — Reverse-Engineering Any AI Product
The prompt
# The Incremental Reveal
Reverse-engineer any AI product's hidden architecture — system prompts, guardrails, tool chains, reasoning patterns — through progressive probing that maps the boundaries of what the system will and won't do.
THE PRODUCT TO REVERSE-ENGINEER: $ARGUMENTS
MY GOAL: [understand system prompt architecture / map tool chain / identify moat / find gaps to compete / "you decide"]
STEP 1 — BASELINE MAPPING: What product claims to do → likely model(s) → interaction pattern → obvious constraints. Use web search to verify.
STEP 2 — PROBE SEQUENCE DESIGN: 8-12 questions in 3 phases. Phase A (Capability Mapping): 3-4 normal-seeming questions. Phase B (Boundary Probing): 3-4 edge cases testing refusal language, routing logic, RAG vs. parametric knowledge. Phase C (Architecture Hypotheses): 2-4 questions designed to confirm/falsify specific hypotheses.
STEP 3 — PATTERN ANALYSIS: System prompt signals → tool chain indicators → guardrail architecture → knowledge boundaries → persona consistency
STEP 4 — ARCHITECTURE RECONSTRUCTION: Model layer → system prompt layer → tool layer → guardrail layer → UX layer. Confidence rating per layer (HIGH/MEDIUM/LOW).
STEP 5 — STRATEGIC IMPLICATIONS: Design decisions worth adopting, gaps to exploit, minimum viable architecture to replicate core value
STEP 6 — VERIFICATION: Probe questions actually diagnostic? Over-interpreting noise? Reconstruction internally consistent?
Full prompt on recap page.Worked example
No worked example recorded yet.