THE THREE DISCIPLINES AROUND THE MODEL

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] Triage this reported agent failure and tell me whether it is primarily a prompt-engineering problem (instruction wording), a context-engineering problem (what information was in scope), or a harness-engineering problem (loop, tool, permission, or budget logic): [PASTE BUG REPORT, TRANSCRIPT, OR TRACE (A STEP-BY-STEP RECORD OF THE RUN) HERE]. Justify the classification with specific evidence from the material provided, and if more than one discipline is implicated, rank them by how much each contributed. Output format: classification line, evidence bullets, ranked contribution list, and one recommended owner and next step per discipline implicated. [END OF PROMPT] [START OF PROMPT] We are deciding where to invest limited engineering time this sprint: improving prompt wording, improving context curation, or improving harness reliability logic, for this agent: [AGENT DESCRIPTION AND CURRENT METRICS: success rate, cost, latency, top failure modes]. Analyze which lever would most likely move our key metrics, given the data provided, and produce a prioritized one-sprint improvement plan with the specific change, expected effect, and effort estimate for each item. Output format: a ranked table (Change, Lever, Expected Effect, Effort). [END OF PROMPT]