Critic Verdict Parser

Background

The Critic agent outputs a structured verdict that the orchestrator loop parses. The canonical format: VERDICT: PASS or VERDICT: FAIL, optional GAP: lines, optional UNSUPPORTED: lines. The orchestrator needs a parser that's tolerant of LLM formatting drift.

Problem statement

Implement parse_critic_verdict(text). Return a dict with:

• passed: bool - True iff a VERDICT: PASS line appears (case-insensitive).
• gaps: list[str] - items from lines starting with GAP:.
• unsupported: list[str] - items from lines starting with UNSUPPORTED:.

Input

• text - the raw critic output.

Output

• dict with keys passed, gaps, unsupported.

Examples

Input: "VERDICT: PASS\nGAP: The impact section is shallow"
Output: {passed: True, gaps: ["The impact section is shallow"], unsupported: []}

Constraints

• Recognise case-insensitive VERDICT: PASS|FAIL.
• A missing verdict means passed: False.
• Strip the prefix and surrounding whitespace from each gap/unsupported.

Notes

• Strict structure, lenient parsing. The Critic prompt enforces these prefixes; the parser still needs to accept extra whitespace, mixed case, and blank lines.