メニュー
Use when asked to look at or address feedback on a PR. Fetches Copilot feedback from both inline threads and review bodies, skips hidden comments, then evaluates and fixes the valid ones.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/keybase/client --skill address-pr-feedbackコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Use when writing, fixing, or adding e2e flow tests for the Keybase app — desktop (Playwright) or iOS (Appium + WebdriverIO). Covers testID conventions, navigation patterns, common pitfalls, and the two-harness structure.
This skill should be used when the user asks to "compare screenshots", "visual diff", "check for visual regressions", "before and after screenshots", "did the UI change", or mentions comparing the desktop app UI between branches or before/after a change. Also triggered by "take baseline", "take current", or "compare against baseline".
Use when hunting redundant, duplicated or looping RPCs in the Keybase client — after a rate-limit error, a service log that rotates every few minutes, a slow or hot app, or to prove a caching fix actually reduced calls. Covers capturing a clean service log against a locally built service and reading it.
| name | address-pr-feedback |
| description | Use when asked to look at or address feedback on a PR. Fetches Copilot feedback from both inline threads and review bodies, skips hidden comments, then evaluates and fixes the valid ones. |
./.claude/skills/address-pr-feedback/get-copilot-feedback.sh [pr-number]
If no PR number is given, it uses the current branch's PR.
Copilot reports findings in two different places, and the script reads both:
[OUTDATED].Comments generated: 0 and puts
its real findings in the review body instead, under ### Suppressed comments. These never
appear as threads.An empty "Inline review threads" section does NOT mean the review was clean. That is the
trap this skill exists to avoid: a thread-only query returns nothing and the PR looks unreviewed
when Copilot in fact filed a finding in the body. Always read both sections before concluding
there is no feedback. If both are empty, check that Copilot has actually reviewed
(gh pr view <n> --json reviews) rather than assuming it found nothing.
The script deliberately strips Copilot's per-file summary table — it restates the diff and carries no findings. It keeps the headline verdict, the one-line concern beneath it, and the suppressed-comments block.
For each finding:
[OUTDATED] ones this matters most: the diff position moved,
but the underlying issue often still exists — find the corresponding spot and evaluate there.
A body finding cites path:line from an older push; re-locate it by symbol, not line number.Do not post replies to GitHub comment threads. Just fix the code.
A comment is hidden only when its thread is resolved or minimized on GitHub. Those are skipped automatically and need no action. Outdated is not hidden — evaluate those normally.