feat(prompts): add security review dimension to code-quality sub-agent

[ceilf6]

Linked Issue Or Context

Closes #369.

Summary

• Extends buildCodeQualityLlmReviewPrompt's system prompt (packages/core/src/sub-agents/code-quality-prompt.ts) so the code-quality review sub-agent actively flags common web security vulnerabilities — XSS (unsanitized data into innerHTML/dangerouslySetInnerHTML), command/SQL injection, secrets hardcoded in client code, and unsafe eval/dynamic code or unvalidated URL/redirect handling — alongside its existing SDD-constraint and maintainability dimensions.
• Findings reuse the existing CodeQualityIssue shape (rule: "security/<kind>", severity), so there is no schema or subagent-logic change.
• This is the review-layer counterpart to the codegen security discipline in feat(prompts): inject security-engineering discipline into codegen #366: generate-safely + review-for-safety defense in depth.

Learned from the Claude Code agentic prompt's security-as-first-class-review-dimension pattern.

Impact Scope

Prompt-only change to the review sub-agent's system string, plus a focused assertion update in code-quality-prompt.test.ts. No schema, subagent execution logic, rule-based fallback, codegen, or planner changes.

GitNexus Impact Summary

• Risk level: MEDIUM
• Critical skeleton changes: none
• GitNexus impact: impact(buildCodeQualityLlmReviewPrompt, upstream) returned LOW (0 direct callers — invoked dynamically). detect_changes rated MEDIUM because the symbol participates as a step in 3 internal prompt-building processes (Main → TruncateFileContent / BuildSddSummary / SummarizeSharedFacts). The change is confined to the system prompt string content: the function signature, return shape ({ system, userPrompt }), and call edges are unchanged, so those 3 processes' contracts are unaffected — the only behavioral delta is that the review LLM now also reports security findings.
• Verification: detect_changes (scope=unstaged) reported changed_count=1 (buildCodeQualityLlmReviewPrompt), affected_count=3 (the prompt-building processes above), risk_level=medium.

Verification

• pnpm --filter @frontagent/core test -- code-quality-prompt.test.ts — 2/2 pass (updated exact-system assertion + security-phrase checks).
• pnpm exec biome check on the 2 changed files — clean.
• pnpm quality:precommit (lint + typecheck + test + test:workflows) — green (25 turbo tasks successful, workflows 32/32).
• gitnexus impact + gitnexus detect_changes — see summary above.

Checklist

• I have linked an issue or explained why this PR stands alone.
• I have kept the diff focused on the stated change.
• I have run pnpm quality:precommit, or explained why it could not run.
• I have run pnpm quality:local for critical skeleton changes, or explained why it could not run. (Prompt-string-only; precommit + GitNexus impact/detect_changes cover it; contract to the 3 processes unchanged.)
• I have updated docs or tests when behavior, public APIs, or Harness contracts changed.
• For critical skeleton changes, I have filled the GitNexus impact summary with concrete results.

[github-actions]

🛡️ ceilf6/repo-guard

代码评审报告: feat(prompts): add security review dimension to code-quality sub-agent

风险等级: 中
处理建议: 批准
决策摘要: ** 可以合并；变更满足 #369 的 prompt-only 验收目标，未改变函数签名、返回结构或下游契约，测试覆盖了新增安全维度与既有语义。

级联分析

• 变更符号: 原始模型未提供结构化级联字段。
• 受影响流程: 原始模型未提供结构化级联字段。
• 变更集外调用方: unknown
• 置信度: degraded

问题发现

模型未返回可结构化的问题发现；已提取可用的决策字段，原始非契约内容未附在评论中。

行级发现

• 无明确变更行归属。

Karpathy 评审

• 假设: 模型输出需要归一化为固定 Markdown 契约。
• 简洁性: 已提取 summary、finding、evidence 与 fix；原始 prose 不再附在评论中，避免占用下游解析与代理上下文。
• 变更范围: 原始模型未提供结构化范围字段。
• 验证: 需要查看 CI、测试或人工 CR 证据补强合并信心。

缺失覆盖

• 输出未命中 Repo Guard Markdown 契约；建议补充真实模型质量评估覆盖。