Files
breadcrumb-the-shire/.agents/prompts/reviewer-code.md

42 lines
1.5 KiB
Markdown
Raw Normal View History

# Code Reviewer Prompt Baseline
Goal:
- verify code quality, architecture compliance, and convention adherence
Input:
- code diff
- `execution-report.json`
- `plan.json`
Required references:
- `.agents/checks/guard-catalog.json` (guards where `reviewer` = `code`)
- `.agents/checks/quality-gates.json`
Output:
- valid JSON by `.agents/contracts/reviewer-code.schema.json`
Scope — verify these guards:
- GR-CORE-003: request input contract
- GR-CORE-007: OpenAPI sync (if API touched)
- GR-CORE-010: idempotent DB scripts (if schema touched)
- GR-CORE-011: settings in DB (if settings added)
- GR-CORE-012: PRG pattern (if POST actions added/changed)
- GR-TEST-001: tests ship with new logic
- GR-TEST-002: code is testable
- GR-UI-014: UI state completeness (if UI touched)
- GR-UI-LIST: list/grid standards (if lists touched)
- GR-UI-DETAIL: detail page standards (if detail pages touched)
- GR-UI-A11Y: accessibility (if UI touched)
- GR-UI-I18N: i18n completeness (if visible text touched)
- GR-UI-REUSE: reuse-first (if JS/CSS added)
- GR-UI-TAXONOMY: status taxonomy consistency (if status enums/labels touched)
- GR-CORE-LAYERS: architecture layer isolation
- GR-CORE-META: agent and codex system integrity (if .agents/ touched)
- GR-CORE-MODULE: module boundary isolation (if modules touched)
Rules:
- findings MUST reference a guard or gate ID
- findings MUST include file path
- use `fail` only when changes are required before finalize
- skip guards not relevant to the change (e.g. skip UI guards for backend-only changes)