Hacktoberfest 2026:維護者為十月標記出來的 issue,仍然開放、適合新手。 瀏覽 Hacktoberfest issue

a failed landing with its work kept asks forever, long after nothing is waiting on the answer

未關閉
#1,328 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

維護者通常 1 天內回覆

還沒有人認領這個 Issue。

評估

難度
5/5
預估耗時
一週以上
新手友好度
42/100
Issue 類型
缺陷
描述清晰度
基本清楚
活躍度
活躍
技術堆疊
go
領域
cli, devtools

研究方向

Start at internal/session/task_status.go:441 and trace how TaskFailed with Held facts becomes ProjectTask status and contributes to needs you. Compare the recorded later nodes, branch existence, and continuation evidence to decide when the conversation has moved past a held landing. Done means the card retains its branch and offer while no longer marking or counting a non-blocking demand.

由索引模型根據 Issue 內容生成。

描述

area:session bug sev:serious

What happens

A task that ends failed with its work still held keeps wearing your call and keeps counting toward needs you, for the rest of the session, with no way to reach zero except answering each one.

internal/session/task_status.go:441:

case TaskFailed:
	if facts.Held {
		// THE ANSWER IS STILL THERE AND SOMEBODY MAY STILL WANT IT...
		status.Presence, status.On = TaskPresenceNeedsLook, TaskWaitPerson
		return status
	}

The offer is real — the branch is on disk and taking it anyway is a genuine answer. The problem is that the offer never expires.

What it looks like

One session, 23 nodes. Four rows stand in needs you, all state=failed, ending=refused, merge=aborted, each with a kept branch that exists:

node title branch
4 Skills: five-part record and store serving task/skills-five-part-record-and-stor-…
6 Skills: chat shelf and use_skill tool task/skills-chat-shelf-and-use-skill-…
7 Store skill fact fields task/store-skill-fact-fields-…
15 File GitHub issue: standing firing sees empty task board task/file-github-issue-standing-firin-…

By the time they were read, the conversation had moved two tranches past them: a later node had already landed a continuation of the same store-record branch, none of the four branches had ever been pushed, and nothing in flight was waiting on any of them. The count said four people-shaped questions. Zero of them were blocking anything.

Why it is wrong

A mark exists only when a person's answer is the one thing unblocking the work. A card may keep showing a state; a mark must not. A mark is a demand and a card is a status. Four standing demands that unblock nothing train a person to stop reading the count, which costs the marks that do matter.

Shape of the fix

A held landing is a question while the conversation is still on it, and a status afterwards. When the work has been superseded, or the session has moved past it, the row keeps its branch and its offer on the card and stops wearing the mark and stops counting toward needs you.

Open question for whoever takes it: what "moved past it" should be measured against. Candidates seen here — a later node that landed the same branch, a node whose branch no longer exists, and plain distance in the graph. The first two are facts already on the record; the third is a guess. Pick from the record, not the clock.

Not this

The four rows are not stale state, not a cache, and not a surface drawing something the engine did not say. This was checked: the engine was restarted onto the current build first, the records were read off disk, and the rows are what ProjectTask returns for those facts. The reading is correct. The demand is what is wrong.

主要語言
Go
星號
115
分支
14
平均合併
9 小時 37 分鐘
30 天內合併 PR
755

環境準備

我們還沒有檢查這個專案的環境設定檔。先看它的 README,通用步驟見我們的新手貢獻指南。

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

Agent-Field/CodeAF 的其他 Issue

查看 Agent-Field/CodeAF 的全部 Issue

相似的 Issue

更多 Go Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。