CONTROL STRUCTURES AND DECISION ROUTING

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] Model this process as a directed graph of nodes and edges (text form), including at least one conditional branch and one retry/guard loop: [PROCESS DESCRIPTION, e.g. "receive resume, extract skills, match to job posting, draft report"]. For each edge, state whether it should be deterministic (hard-coded) or model-driven (left to the agent's judgment), and justify each from a safety/reliability standpoint. Then review this existing routing code for any edge that is model-driven but should be deterministic: [PASTE LAB routing/graph CODE HERE]. Output format: graph description, edge-type table, then a flagged-edge list with fixes. [END OF PROMPT] [START OF PROMPT] For each of these decision points in our agent, tell me whether it should be deterministic or model-driven, and why: [LIST DECISION POINTS, e.g. "whether to retry a failed tool call", "whether a resume matches a job posting", "whether to escalate to a human"]. Where you recommend model-driven, specify what guardrail should still bound the model's choice. Output format: a table (Decision Point, Recommendation, Guardrail, Justification). [END OF PROMPT]