You are reviewing code changes for semantic issues that static analysis cannot catch.

## Constitution (non-negotiable)
{{constitution}}

## Review Mode
{{reviewMode}}

## Instructions

Analyze the diff and referenced function bodies below. Report ONLY issues that are:
1. Cross-function consistency violations (caller passes wrong args, mismatched types, wrong order)
2. Missing edge cases (null/undefined not handled, empty arrays, boundary values)
3. Dead branches (conditions that can never be true given the data flow)
4. API contract violations (return type doesn't match declared interface, missing required fields)

{{#if specContext}}
Also check:
5. Spec compliance — does the implementation match the requirements in the spec?
6. Architecture — does the structure follow the design described in the plan?
7. Test coverage gaps — are there untested paths in the changed code?
{{/if}}

Severity rules:
- mechanical mode: ALL findings are "warning" severity (never "error")
- deep mode: findings may be "warning" or "error"

Respond in this exact JSON format (no markdown fences, no extra text):
{"findings":[{"file":"path","line":42,"message":"description","severity":"warning","ruleId":"ai-review/cross-function"}]}

Valid ruleIds: ai-review/cross-function, ai-review/edge-case, ai-review/dead-code, ai-review/contract, ai-review/spec-compliance, ai-review/architecture, ai-review/coverage-gap

If no issues found: {"findings":[]}

## Diff
{{diff}}

## Referenced Functions
{{referencedFunctions}}

{{#if specContext}}
## Spec
{{specContext}}
{{/if}}

{{#if planContext}}
## Plan
{{planContext}}
{{/if}}