Agents 参考
上游 spec-first 当前提供 26 个 agents，它们是 skill-local prompt 资产，分布在 skills/**/references/agents/ 下（共 46 处调用，跨 skill 复用）。Claude Code 与 Codex 都会在 spec-first init 时把它们随所属 skill 一起生成为 host-specific runtime copies；源码与运行时副本需要分开理解。
主路径：先打开对应 Skill 详情页 的 Agent / Persona 列表，再下钻 agent 详情。本页是跨 skill 汇总索引，方便全局检索与对照。
Agents 是 workflow 可调用的专家能力，不是一次任务必须全部启动的固定队列。Spec-First 的调度原则是：先判断当前研发阶段，再根据文档内容、diff 文件、风险域、技术栈、证据缺口和宿主能力选择最小必要专家。
Agent Dispatch
Agent 不是常驻队列，而是 workflow 在证据缺口或风险域明确时调用的专家视角。
Doc Review需求、计划和任务包审查coherence、feasibility、product、design、security、scope 等 persona 只在文档风险需要时进入。
Code Review按 diff 和风险域选择 revieweralways-on reviewer 保底，API、migration、performance、Swift、Rails 等条件 reviewer 按文件和语义触发。
Research补足外部、历史和组织上下文repo、git history、sessions、Slack、issues、framework docs 和 best practices agent 提供证据，不替代最终取舍。
SpecialistUI、部署、安全和数据专项判断专项 agent 聚焦局部风险，把发现回传给上层 workflow 汇总。
资产边界
| 层级 | 位置 | 当前核验 |
|---|---|---|
| Source agents | 上游 skills/**/references/agents/*.md | 26 个唯一 agent（46 处 skill-local 调用，无 frontmatter） |
| Claude runtime agents | .claude/agents | 由 spec-first init 从 source 生成的 host runtime copies |
| Codex runtime agents | .codex/agents | 由 spec-first init 从 source 生成的 host runtime copies |
按 Skill 归属的 Agent 索引
下面的索引由 npm run agents:pages 基于上游 skill-local skills/**/references/agents/*.md 生成：先按 归属 skill 分组，再列出该 skill 挂载的 agents（跨 skill 复用会出现在多个分组）。这些 prompt 资产没有 frontmatter，详情页的 description 由生成器从 prompt 首段启发式提取；每个详情页说明归属 skill、复用次数、输入输出、边界和 source/runtime 关系。
主发现路径仍在各 Skill 详情页 的 Agent / Persona 列表（含 persona）；本页只做 agents 的跨 skill 汇总。
spec-brainstorm
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-slack-researcher | Research | You are an expert organizational knowledge researcher specializing in extracting actionable context from Slack conversations. |
spec-compound
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-best-practices-researcher | Research | You are an expert technology researcher specializing in discovering, analyzing, and synthesizing best practices from authoritative sources. |
spec-data-integrity-guardian | Release / Data Safety | You are a Data Integrity Guardian, an expert in database design, data migration safety, and data governance. |
spec-framework-docs-researcher | Research | You are a meticulous Framework Documentation Researcher specializing in gathering comprehensive technical documentation and best practices for software libraries and frameworks. |
spec-pattern-recognition-specialist | Architecture | You are a Code Pattern Analysis Expert specializing in identifying design patterns, anti-patterns, and code quality issues across codebases. |
spec-performance-oracle | Security / Performance | You are the Performance Oracle, an elite performance optimization expert specializing in identifying and resolving performance bottlenecks in software systems. |
spec-security-sentinel | Security / Performance | You are an elite Application Security Specialist with deep expertise in identifying and mitigating security vulnerabilities. |
spec-session-historian | Research | You are an expert at extracting institutional knowledge from coding agent session history. |
spec-explain
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-work-recap-scout | Research | You are a work-recap scout. |
spec-ideate
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-issue-intelligence-analyst | Research | You are an expert issue intelligence analyst specializing in extracting strategic signal from noisy issue trackers. |
spec-learnings-researcher | Research | You are a domain-agnostic institutional knowledge researcher. |
spec-slack-researcher | Research | You are an expert organizational knowledge researcher specializing in extracting actionable context from Slack conversations. |
spec-web-researcher | Research | You are an expert web researcher specializing in turning open-ended search queries into a focused, structured external grounding digest. |
spec-optimize
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-learnings-researcher | Research | You are a domain-agnostic institutional knowledge researcher. |
spec-repo-research-analyst | Research | You are an expert repository research analyst specializing in understanding codebases, documentation structures, and project conventions. |
spec-plan
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-agent-native-planning-strategist | Architecture | You are an agent-native planning strategist. |
spec-architecture-strategist | Architecture | You are a System Architecture Expert specializing in analyzing code changes and system design decisions. |
spec-best-practices-researcher | Research | You are an expert technology researcher specializing in discovering, analyzing, and synthesizing best practices from authoritative sources. |
spec-data-integrity-guardian | Release / Data Safety | You are a Data Integrity Guardian, an expert in database design, data migration safety, and data governance. |
spec-data-migration-reviewer | Release / Data Safety | You are a data migration and schema-change reviewer. |
spec-deployment-verification-agent | Release / Data Safety | You are a Deployment Verification Agent. |
spec-flow-analyzer | Research | Analyze specifications, plans, and feature descriptions from the end user's perspective. |
spec-framework-docs-researcher | Research | You are a meticulous Framework Documentation Researcher specializing in gathering comprehensive technical documentation and best practices for software libraries and frameworks. |
spec-git-history-analyzer | Research | You are a Git History Analyzer, an expert in archaeological analysis of code repositories. |
spec-learnings-researcher | Research | You are a domain-agnostic institutional knowledge researcher. |
spec-pattern-recognition-specialist | Architecture | You are a Code Pattern Analysis Expert specializing in identifying design patterns, anti-patterns, and code quality issues across codebases. |
spec-performance-oracle | Security / Performance | You are the Performance Oracle, an elite performance optimization expert specializing in identifying and resolving performance bottlenecks in software systems. |
spec-repo-research-analyst | Research | You are an expert repository research analyst specializing in understanding codebases, documentation structures, and project conventions. |
spec-security-sentinel | Security / Performance | You are an elite Application Security Specialist with deep expertise in identifying and mitigating security vulnerabilities. |
spec-slack-researcher | Research | You are an expert organizational knowledge researcher specializing in extracting actionable context from Slack conversations. |
spec-web-researcher | Research | You are an expert web researcher specializing in turning open-ended search queries into a focused, structured external grounding digest. |
spec-pov
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-external-evidence-researcher | Research | You are an external-evidence researcher for a verdict skill. |
spec-pov-peer | Specialist | Form an independent, decisive point of view on the supplied subject, grounding yourself in evidence you inspect in the shared working tree. |
spec-precedent-activity-scout | Research | You are a precedent-&-activity scout for a verdict skill. |
spec-project-grounding-scout | Research | You are a project-grounding scout for a verdict skill. |
spec-resolve-pr-feedback
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-pr-comment-resolver | Documentation / PR | You implement one PR review fix that the orchestrator has already judged valid and worth doing. |
spec-sweep
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-media-analyzer | Research | You are a media-analysis specialist inside an already-running spec-sweep pass. |
spec-work
| Agent | 分组 | 主要能力 |
|---|---|---|
spec-figma-design-sync | UI / Design | You are an expert design-to-code synchronization specialist with deep expertise in visual design systems, web development, CSS/Tailwind styling, and automated quality assurance. |
调度总图
text
Spec / Plan document
|
v
spec-doc-review
|
+-- always-on personas: coherence + feasibility
+-- conditional personas: product / design / security / scope / adversarial
Code diff / PR
|
v
spec-code-review
|
+-- always-on personas: correctness + testing + maintainability + project-standards
+-- conditional personas: security / performance / API / migration / reliability / 语言栈
Research / planning / debugging / knowledge
|
v
workflow-selected agents (skill-local references/agents/)
|
+-- repo-profiler / git-history / sessions / Slack / issues / framework docs / best practices
UI / deployment / data safety
|
v
specialist agents
|
+-- figma-design-sync / deployment-verification / data-integrity / data-migrationAgents 与 Personas 的区别
上游把 skill-local prompt 资产分成两类，都位于对应 skill 的 references/ 目录下，都没有 frontmatter，都由 spec-first init 随 skill 一起生成，不是用户直接调用的 spec-* 入口：
- Agents（
skills/**/references/agents/*.md，26 个唯一 / 46 处调用）：research、grounding、architecture、security、performance、UI、PR 等专家判断角色，即上方「按 Skill 归属的 Agent 索引」列出的清单。 - Personas（
skills/**/references/personas/*.md，26 个）：spec-code-review、spec-doc-review、spec-simplify-code的内置 reviewer 视角，按文档/diff 风险域触发；列在所属 skill 详情页的 Agent / Persona 列表，不单独生成全局详情页。
按研发阶段理解 agents
| 阶段 | 常见 agent 能力 | 典型触发 |
|---|---|---|
| Codebase | repo research、git history、sessions、learnings | 新项目接入、计划前 research、调试历史问题、复用过往经验 |
| Provider Evidence | provider 通常不是 agent；agent 负责解释和补证据 | provider degraded 后需要 bounded direct reads 或历史解释 |
| Spec | spec-flow-analyzer、external-evidence-researcher、project-grounding-scout、web / issue research | 需求不清、用户流缺失、外部 prior art 或项目 grounding 未确认 |
| Plan | architecture-strategist、agent-native-planning-strategist、framework-docs-researcher、best-practices-researcher、doc-review personas | 计划要落地、技术方案需要审查、依赖或架构风险较高 |
| Tasks | doc-review personas 复核 task pack 是否忠于 source plan | task pack 可能扩 scope、依赖不清、同 wave 文件重叠 |
| Code | worker/explorer 子任务、stack specialists、design / performance / security specialists | 执行实现、调试、优化、UI 打磨 |
| Review | always-on reviewers + conditional reviewers | PR / diff 质量门 |
| Knowledge | session historian、Slack researcher、learnings researcher、related docs finder 类能力 | 复盘、知识沉淀、刷新 stale docs |
事实来源
本页以上游 skill-local agent scan 交叉核验的 26 个唯一 agents（46 处调用）为准。数量事实由官网 spec-first-facts.json 与内容审计脚本校验。
阅读下一步
- Skills 参考：35 个 source skill；各详情页含 Agent / Persona 列表
- Code Review 指南：always-on 与条件 reviewer 如何调度
- 工作流总览：agents 在研发链路中的位置
spec-code-review详情：查看该 skill 的 Agent / Persona 列表与 always-on personas