Planner Output Parser

Background

A research / planning agent often outputs a numbered list of sub-questions. The downstream agent loop needs a robust parser: handle numbering styles (1., 1), 1:), strip Markdown emphasis, ignore blank lines, trim whitespace.

Problem statement

Implement parse_plan(text). Walk lines of text; for each line matching a numbered-list pattern, extract the content. Return a list[str] of cleaned sub-questions in order.

Input

• text - the raw LLM output containing the plan.

Output

• list[str] of cleaned sub-questions; empty list if no items match.

Examples

Input: "1. What is X?\n2) How does Y work?\n3: Why Z?"
Output: ["What is X?", "How does Y work?", "Why Z?"]

Constraints

• Accept N., N), N: (single digit or multi-digit).
• Strip leading/trailing whitespace and Markdown **bold** / *italic* around the whole item.
• Skip lines that don't match the pattern.

Notes

• Why robust parsing. LLMs produce inconsistent formatting; brittle parsers fail silently and the planner returns 0 sub-questions instead of erroring.