df
GitHub用于处理代码审查反馈,自动检查差异状态、读取评论并回复,解决或标记已处理意见,支持等待新反馈及定位代码位置。
触发场景
安装
npx skills add matheusfillipe/diffler --skill df -g -y
SKILL.md
Frontmatter
{
"name": "df",
"description": "Check the diffler review and respond to the human's feedback. Use when the user says \"check diffler\", invokes \/df, or asks to look at review comments."
}
Check the diffler review and respond to the human's feedback:
- Call the diffler MCP tool
review_statusfor the active review and its changed files. Ifwalkthroughsholds none about a change you made, callpublish_walkthroughfirst (see /dfa), then continue.review_statusnames therepo; if it is not the repository you mean, calllist_instances, thenuse_instancewith that repo, and start over. If you changed files in another git repository, callopen_projectwith its path, so the human reviews it as a tab beside this one;projectslists the ones already open. - Call
get_commentswith status "open" and read each comment in place. Each carries asource: the human's review comments on the diff itself carry the diff's own source, while one startingwalkthrough-is feedback on that walkthrough, a reply on one of its stops or notes. - Address every comment in the code it anchors to.
- Answer each with
reply_comment(what you changed and why), thenpropose_resolveto flag it addressed. That flag adds no text: do not summarise the answer you just gave. Only the human resolves for real, in the TUI. - Call
wait_for_feedbackwith the latest epoch and start over when it returns: the human just sent new feedback. It answers within 55 seconds; atimed_outresult means they are still reviewing, so call it again. If the call itself fails, checkreview_status: diffler is closed only when that fails too, otherwise keep waiting.
Every diffler tool call already shows in the human's status bar. Before a
stretch of work no tool call covers, such as editing the code a comment asks
about, call report_activity with a few words for it.
When the human asks where something is, call focus to take them there:
a comment or a walkthrough by its id, or a file and line in any review.
Write every reply like this
- Say what you changed and why, in two to four short sentences or bullets. Lead with the change, then the reason.
- Answer the question that was asked. If you disagree, say so in one sentence and say why.
Write for the card
Write for a narrow card beside the code:
- Full sentences with their articles and pronouns, in the first person plural
when we are the subject: "We changed X because Y." A sentence may also lead
with the thing itself: "
infraruns first because Y." Do not open every sentence the same way. Cutting words does not make text clearer. - Put every identifier in backticks:
compute_overrides,SHEET_PERMISSIONS,MissingDecision. The card renders markdown; bare identifiers read as prose. - Plain verbs for code: raise, return, check, read, write, fail, skip. No metaphor: no gate, carry, land, surface, shape, vote, mint, unpick, wall, hold an opinion.
- Comparing things, or listing data with several fields per item: a markdown table, header row, two to four columns, one row per thing, identifiers in backticks inside cells. The card rules the columns.
If the diffler tools are missing, the TUI isn't running: ask the human to
run diffler in the repository first.
版本历史
- 78ce09e 当前 2026-11-21 03:49


