Handle unknown disposition render values

[wwind123]

Summary

• add a defensive fallback when rendering prior item dispositions so new normalized values do not raise KeyError
• add regression coverage for unknown disposition rendering through the public review comment path

Fixes #142

[wwind123]

Implemented the fix for issue #142 in PR #163. _render_disposition_status now falls back to the raw normalized disposition string instead of raising KeyError, and the regression test covers that path through public review comment rendering.

Tests: python3 -m pytest tests/test_agent_loop.py -k "render_public_review_comment or unresolved_item_dispositions" (passed); python3 -m pytest (passed)

-- OpenAI Codex

[wwind123]

Code Review — PR #163: Handle unknown disposition render values

Fixes: #142

Summary

This PR applies the defensive fallback suggested in issue #142 to _render_disposition_status, changing the plain dict access to .get() with the raw disposition string as the default. The change is minimal, targeted, and correct.

Change analysis

orchestrator.py — _render_disposition_status

# Before
rendered = labels[disposition.disposition]

# After
rendered = labels.get(disposition.disposition, disposition.disposition)

The fix is correct. _normalize_disposition in protocol.py currently raises AgentLoopError for any value outside the known set (resolved, blocking, same-pr, same-plan, future), so under today's parsing path the KeyError is unreachable. However, the ReviewItemDisposition dataclass can also be constructed directly (e.g., deserialized from stored round metadata via _deserialize_disposition) or built in tests, so the defensive fallback provides genuine safety at the extension boundary identified in #142. Surfacing the raw normalized string is the right default — it's informative rather than silent.

tests/test_agent_loop.py — test_render_public_review_comment_preserves_unknown_disposition_values

The test correctly exercises the new code path end-to-end through _render_public_review_comment, constructing a ReviewItemDisposition with disposition="deferred" directly (bypassing the parser, which is the exact gap the fix guards). The assertion verifies that the rendered output passes through "deferred: tracked for a later parser update" verbatim rather than raising. This is exactly the regression coverage the issue called for.

CI

GitHub checks are pending (test job in_progress) at review time. The author reports a full local pytest run passed. The logic of the change is straightforward enough that CI result is unlikely to diverge, but this review is noted against a pending CI state.

No blocking issues found

The fix precisely matches the recommendation in issue #142, is correctly implemented, and is covered by a well-scoped regression test.

-- Anthropic Claude

[wwind123]

The changes in this PR correctly address the risk of a KeyError in _render_disposition_status by using .get() with a fallback to the raw disposition value. This follows the recommendation from issue #142 to provide a defensive fallback for unrecognised disposition values, making the codebase more resilient to future extensions of the disposition set.

Changes:

• src/coding_review_agent_loop/orchestrator.py: Updated _render_disposition_status to use labels.get(disposition.disposition, disposition.disposition).
• tests/test_agent_loop.py: Added test_render_public_review_comment_preserves_unknown_disposition_values to verify that unknown dispositions are rendered correctly using the raw value as a fallback.

Verification:

• Ran the new test case: test_render_public_review_comment_preserves_unknown_disposition_values passed.
• Ran all tests in tests/test_agent_loop.py: All 345 tests passed.

The implementation is surgical, well-tested, and aligns with the project's existing patterns.

-- Google Gemini

[wwind123]

GitHub PR checks are still pending for PR #163.

Reviewer approvals do not make this PR merge-ready until GitHub PR checks are green, or the PR explicitly states that only a local subset passed.

• Pending checks: test (in_progress)
• Required status checks are not configured on the PR base branch.

-- coding-review-agent-loop