Fix is_clarification_request to defer to final state marker over embedded AGENT_CLARIFY

[wwind123]

Summary

• is_clarification_request() previously returned True for any response containing <!-- AGENT_CLARIFY --> anywhere, including examples in prose or code blocks.
• A plan ending with <!-- AGENT_PLAN_STATE: blocking --> but containing a literal AGENT_CLARIFY example in its body would incorrectly abort the loop as a clarification request.
• Fix: when a valid AGENT_STATE, AGENT_PLAN_STATE, or AGENT_PR marker appears after the last AGENT_CLARIFY in the text, the state marker takes precedence and the response is not treated as a clarification.

Fixes #278

Test plan

• test_is_clarification_request_state_marker_after_clarify_wins: plan/PR/coder body with embedded AGENT_CLARIFY example followed by a valid footer → not clarification.
• test_is_clarification_request_clarify_as_final_marker_is_still_detected: AGENT_CLARIFY as the final marker → still a clarification; AGENT_STATE in body then AGENT_CLARIFY at end → clarification wins.
• test_is_clarification_request_code_block_example_does_not_trigger: AGENT_CLARIFY in a fenced code block + AGENT_PLAN_STATE footer → not clarification.
• All 710 existing tests continue to pass.

🤖 Generated with Claude Code

[wwind123]

Fix: AGENT_CLARIFY no longer overrides a valid final state marker

Root cause

is_clarification_request() in protocol.py called CLARIFY_RE.search(text), which matched any occurrence of <!-- AGENT_CLARIFY --> anywhere in the response — including literal examples in plan prose or code blocks describing the protocol.

Fix

is_clarification_request() now checks whether a valid AGENT_STATE, AGENT_PLAN_STATE, or AGENT_PR marker appears after the last AGENT_CLARIFY match. If one does, that final state marker takes precedence and the response is not treated as a clarification request.

def is_clarification_request(text: str) -> bool:
    matches = list(CLARIFY_RE.finditer(text))
    if not matches:
        return False
    last_clarify_pos = matches[-1].end()
    remaining = text[last_clarify_pos:]
    if STATE_RE.search(remaining) or PLAN_STATE_RE.search(remaining) or PR_RE.search(remaining):
        return False
    return True

Acceptance criteria coverage

• Plan with embedded AGENT_CLARIFY example + ending AGENT_PLAN_STATE: blocking → handled as blocking ✓
• Coder/PR response with embedded AGENT_CLARIFY example + ending AGENT_STATE: blocking or AGENT_PR: N → not treated as clarification ✓
• Response whose final marker is AGENT_CLARIFY → still treated as clarification ✓
• AGENT_STATE in body then AGENT_CLARIFY at end → clarification wins ✓

All 710 existing tests pass; 6 new tests added covering issue #278 shapes.

-- Anthropic Claude

[wwind123]

Accidental duplicate of PR #279