Agent Persona Exploration - 2026-10-29
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 2/5
- 预计耗时
- 1-3 小时
- 新手友好度
- 76/100
调研方向
从 .github/aw/create-agentic-workflow.md 开始,这是三条建议中都提到的文件。阅读其中有关网络权限、假设和指导的章节,然后根据此 issue 中的发现添加建议的默认设置和决策指导。完成标准是文档涵盖仅使用制品的工作流、常见的基线或定义歧义,以及引用所查阅指导文件的说明。
由索引模型根据 Issue 内容生成。
描述
Persona Overview
- Agent: agentic-workflows (gh-aw custom agent)
- Personas This Run: Frontend Developer, DevOps Engineer, QA Tester
- Scenarios Tested: 4 (FD-2, DO-1, DO-2, QA-1), out of 6 generated
- Average Quality Score: 4.6/5.0
This completes the full 9-persona rotation cycle (prior runs covered Program Manager, Designer, Information Worker, Legal / Compliance, Product Manager, Backend Engineer).
Key Findings
- All 4 invocations succeeded; no misunderstandings of task intent observed.
- The agent consistently reaches for
github(gh-proxy, default/actions toolsets) and skipsplaywright/web-fetchwhen the task is pure repo/Actions-API analysis — tool selection stayed tight across all scenarios. - Triggers were well-matched to the data source:
workflow_runto react to an existing build instead of duplicating it (FD-2), path-scopedpull_requestfor workflow-file review (DO-2),schedule+workflow_dispatchfor the digest (DO-1). - Every scenario included a
noopbranch for the "nothing to report" case, and read-heavy scenarios converged onadd-commentwithhide-older-comments: truefor rolling per-PR updates vs.create-issuewithclose-older-issues: truefor periodic digests. - One scenario (QA-1) loosened network access to
defaultsinstead ofnone/[], breaking from the pattern the other three followed for artifact-only reads.
Top Patterns
- Triggers: scoped
pull_request(path-filtered or type-filtered),workflow_runreacting to a build,schedule+workflow_dispatchfor periodic digests. - Tools:
github(gh-proxy) as the near-universal default;actionstoolset added only when artifact/run-history access is actually needed. - Security: read-only permissions scoped to exactly what's needed (
contents,pull-requests,actions— never broader); network access minimized to none for artifact-only workflows in 3/4 cases.
View High Quality Responses (Top 2-3)
- DO-2 (5.0/5) — Workflow-file security review. Path-scoped trigger (
.github/workflows/**), zero network access, and a prompt that directly operationalizes the security checks it was asked to automate (SHA-pinning, permissions-block scoping). Cited the specific.github/awdoc paths it drew guidance from. - FD-2 / DO-1 (4.6/5 each) — Both named an explicit unresolved assumption (artifact baseline source; definition of "flaky") with a stated fallback instead of silently guessing, then asked for confirmation.
View Areas for Improvement (Top 2-3)
- QA-1 proposed
network: defaultsfor a workflow that only reads CI artifacts via thegithubtool — the other three scenarios in this run converged on fully closed network access (none/[]) for equivalent artifact-only reads, so this is an inconsistency rather than a necessity. - Baseline/definition assumptions (bundle-size baseline source, flakiness definition) are consistently flagged but not resolved with a stated default — good transparency, but each one currently requires a follow-up round-trip with the user.
Recommendations
- Add explicit guidance in
.github/aw/create-agentic-workflow.mdthat artifact-only / Actions-API-only workflows should default tonetwork: allowed: []rather thannetwork: defaults, to make the pattern observed in 3 of 4 scenarios the stated default rather than an emergent convention. - Extend
.github/aw/create-agentic-workflow.mdwith a short decision-tree for common "baseline/definition" ambiguities (e.g., bundle-size baseline source, flakiness definition window) so the agent can propose a sensible default inline instead of only flagging the assumption. - Document the "cite the specific guidance files consulted" behavior seen in DO-2 as a recommended pattern in
.github/aw/create-agentic-workflow.md, since it improves response auditability without added verbosity.
References:
Generated by 🎭 Agent Persona Explorer · claude · agent · ~103 AIC · ⌖ ~1 AIC · ⊞ 4.5K · ◷
- 主要语言
- Go
- 星标
- 5.5k
- 派生
- 606
- 平均合并
- 4 小时 35 分钟
- 30 天内合并 PR
- 807
环境准备
在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。
- 提供 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
github/gh-aw 的其他 Issue
-
automation doc-healer documentation
难度 2/5 1-3 小时 新手友好度 65/100
维护者通常 1 天内回复
-
agent-research
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
automation cookie
难度 2/5 1-3 小时 新手友好度 68/100
维护者通常 1 天内回复
-
agentic-workflows automation aw-essential cookie
难度 2/5 1-3 小时 新手友好度 75/100
维护者通常 1 天内回复
-
automated-analysis code-quality cookie refactoring
难度 2/5 1-3 小时 新手友好度 65/100
维护者通常 1 天内回复
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 90/100
tsukumogami/tsuku#2804 ·
维护者通常 1 天内回复
-
Anthropic /v1/messages: only the last thinking block of an assistant message is kept可能已有人在做 @Shayan12456 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 88/100
维护者通常 1 天内回复
-
raised-by:verifier
难度 2/5 1-3 小时 新手友好度 82/100
medici-finance/assay#5039 ·
维护者通常 1 天内回复
-
metax CheckHealth reads Status.Capacity, so unhealthy Metax-GPU nodes are never cleaned up可能已有人在做 @barbierajput378-pixel 今天认领。 未关闭kind/bug
难度 2/5 1-3 小时 新手友好度 82/100
Project-HAMi/HAMi#3343 · 1 条评论 ·
维护者通常 1 天内回复
-
area:recipes event:contribfest format:human-first source:agent status:needs-triage type:recipe
难度 2/5 1-3 小时 新手友好度 85/100
cadence-workflow/cadence-ai-samples#140 ·
维护者通常 1 天内回复