Agent Persona Exploration - 2026-11-10

オープン 初心者向け
  • 何か問題がありますか？
#76,552 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
2/5
見積もり時間
1〜3時間
初心者へのやさしさ
69/100
issue の種類
ドキュメント
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
github-actions, go
領域
ai-infra-agents, documentation

調査の方向性

まず、指定された3つのガイド .github/aw/safe-outputs-content.md、.github/aw/create-agentic-workflow-trigger-details.md、.github/aw/create-agentic-workflow.md を読んでください。トリガーのガイダンスを編集する前に、GitHub が文書化している milestoned イベントの動作を確認してください。ガイドで、明示的な目標数、マイルストーンの注意点とガード、未定義の用語についてデフォルトと明記したデフォルト案を扱ったら完了です。

索引モデルが issue の本文から書いたものです。

説明

agent-research cookie
Persona Overview
  • Agent: agentic-workflows
  • Personas This Run: Program Manager, Designer, Legal / Compliance
  • Scenarios Tested: 3 of 6 generated (PGM-4, DSG-4, LEG-6)
  • Average Quality Score: 4.3/5.0 (3 of 3 invocations succeeded)
Key Findings
  • All three invocations succeeded. Each returned one recommendation covering trigger, tools, permissions, network access, and safe outputs.
  • Every recommendation used deny-by-default network (allowed: []), read-only contents, no edit tool, and allowlisted labels or reviewers.
  • Prompt clarity was the lowest dimension (4.0). Ambiguous terms (freeze-date format, what counts as uncertain, tracker location) were left as open questions rather than given defaults.
  • PGM-4 scored lowest (4.0). The agent itself flagged target: "*" as too broad.
  • The average is slightly below the last two runs (4.6 and 4.4). Three scenarios are too few to call that a trend.
Top Patterns
  1. Triggers: pull_request with paths: filters for file-scoped checks (DSG-4, LEG-6); issues: milestoned for event-driven milestone checks (PGM-4), matching the decision table in .github/aw/create-agentic-workflow-trigger-details.md.
  2. Tools: the default github toolset plus default bash for static checks. No edit or playwright in any recommendation.
  3. Security: network: allowed: [] in all three, allowlisted add-labels and reviewer teams, and explicit noop branches.
View Scores by Scenario
Scenario Trigger Tools Security Prompt Completeness Average
PGM-4 (Program Manager, issue) 4 4 4 4 4 4.0
DSG-4 (Designer, PR) 5 4 4 4 5 4.4
LEG-6 (Legal / Compliance, PR) 4 5 5 4 4 4.4
View High Quality Responses (Top 2-3)
  • DSG-4 (4.4): Path-filtered pull_request trigger. Explicit allowed values (currentColor, none, inherit, url(#...)) and a defined path for files that cannot be checked (invalid XML, external references, size limit). Says clearly that the checks are structural and do not judge contrast or title quality.
  • LEG-6 (4.4): Routes uncertain cases to a legal reviewer and a label instead of deciding them. Minimal permissions and no network. The per-image table is easy to act on.
View Areas for Improvement (Top 2-3)
  • PGM-4 target: "*": lets the agent comment on any issue. Restricting targets to the triggering issue and the tracker by number would fix it. Its own caveat says the same.
  • PGM-4 milestone read via gh api inside the agent: in this environment gh is not authenticated for reads, so the pattern may fail if copied. The .github/aw/ docs place gh api calls in deterministic steps (for example workflow-patterns.md). The milestone read could move to a pre-agent step. Not verified against the runtime.
  • DSG-4 SVG checks as ad hoc Python: a deterministic pre-agent step, as run 42644982485 recommended for FE-3, would make results reproducible and limit the agent to reading output.
  • Undefined terms: the agent deferred definitions (freeze-date format and timezone, what counts as uncertain) to the user. A proposed default, marked as a default, would be more actionable.
Recommendations
  1. In .github/aw/safe-outputs-content.md, advise against target: "*" when a workflow acts on the triggering item plus one named tracker. List explicit numbers instead.
  2. In .github/aw/create-agentic-workflow-trigger-details.md, add a caveat to the milestone trigger row. The evaluator says milestoned may not fire for issues created directly into a milestone. Check the GitHub docs for the event behaviour before documenting it, then recommend a guard.
  3. In .github/aw/create-agentic-workflow.md, require ad hoc evaluations to propose a default for undefined terms and label it as a default, instead of listing them only as open questions.

References: §42844667323, §42644982485, §42494947077

Generated by 🎭 Agent Persona Explorer · claude · routed: haiku55 xhigh · ~8 AIC · ⌖ ~2 AIC · ⊞ 7.6K · ◷

主要言語
Go
スター
5.6k
フォーク
615
平均マージ
5時間 21分
マージ済み PR（30日）
792

環境構築

Codespaces で開く

このプロジェクトの開発コンテナを、あなたの GitHub アカウントでブラウザ上に起動します。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

github/gh-aw のほかの issue

github/gh-aw の issue をすべて見る

似ている issue

Go の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。