CONTEXT CONSTRUCTION (THE CONTEXT BUILDER)

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS Prompt 1 helps you design the modular prompt and restructure real code. [START OF PROMPT] Design a modular system prompt for this agent, broken into labeled blocks (identity, standing rules, current task data, relevant retrieved context), and specify the ordering that best keeps the model's attention on the most important rules: agent = [AGENT DESCRIPTION, e.g. Resume Review Agent], requirements = [SPECIFIC REQUIREMENTS]. Then show how this existing context-builder code should be restructured to produce that ordering: [PASTE LAB context_builder.py HERE]. Output format: a block-by-block prompt template, then a code-change list. Explain briefly why each block sits where it does. [END OF PROMPT] Prompt 2 helps you fight rule drift with a reminder. [START OF PROMPT] We have observed this agent drifting off a rule in long-running sessions: rule = [RULE, e.g. "never fabricate a skill not present in the resume text"], evidence = [PASTE TRANSCRIPT SHOWING DRIFT HERE]. Write a system-reminder snippet to re-inject periodically, specify the exact re-injection frequency or trigger, and show where in the context builder it should be inserted. Output format: the reminder text, the trigger rule, and the insertion point in the code. Also note whether this rule should additionally be enforced by code rather than by reminder alone. [END OF PROMPT]