Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

The manual has no heading answering "can you use skills?", so the chat invents a denial

オープン
  • 何か問題がありますか？
#1,378 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
3/5
見積もり時間
1〜2日
初心者へのやさしさ
78/100
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
go
領域
documentation, testing-qa

調査の方向性

Start with the probes in internal/manual/chat_test.go and the skill pages putting-a-skill-in-front.md, use-skill.md, and skills-a-turn-used.md. Add a direct capability heading to the appropriate page covering both enabled and switched-off states, then verify the listed probes fail before the heading exists and pass afterward.

索引モデルが issue の本文から書いたものです。

説明

area:chat bug sev:papercut

What happened

Asked can you use skills ? on 2f94ca1f0 — a binary carrying the skills wave
(#1365) and its manual pages — the chat answered:

The manual mentions "installed skills" in passing but doesn't explain

and then, after reaching for list_harnesses and list_subharnesses:

If you meant skills in the Claude Code SKILL.md sense, codeaf has no such
mechanism — harnesses and the notebook are what it has instead.

The program denies a feature it ships.

The gap this issue is about

internal/manual/chat_test.go contains no probe mentioning skills. Grepped
the whole table. So the retrieval gate has never once checked that a person's
first question about the newest feature reaches the pages written for it.

The pages themselves are good — putting-a-skill-in-front.md, use-skill.md,
skills-a-turn-used.md. Their headings are:

# Putting a skill in front of this conversation
## How do I use a skill for this
## Attach a skill from any folder
## How do I turn a skill off
# Skills a turn used
## Which skills did it use?
## Did it use my skill?
# use_skill — list or get a skill from the shelf

Every one assumes the asker already knows skills exist. They answer "how do
I use one" and "which did it use". None answers "do you have this at all",
which is the question a person actually asks first, and the one that failed.

This is the failure the manual law names: people search for the word they have,
not the word the writer used. The existing gates pass, because they check that
every tool name and slash command is mentioned somewhere — and use_skill is
mentioned. Mention is not reachability.

Why a correct answer was unreachable in BOTH states

This matters for the fix, because there are two true answers and the corpus has
neither.

On this machine memory.enabled is off, so config.Memory is nil,
useSkillTool (internal/session/tools_skill.go:71) correctly returns nil, and
the verb is absent — the absent-not-broken law working as designed. The true
answer was therefore "skills are switched off on this machine", not "codeaf
has no such mechanism".

With memory on and a populated shelf, the true answer is "yes, here is how".

The corpus has no heading that answers the capability question in either state,
so the model had nothing to retrieve and improvised. A page that cannot be
reached by the question it answers is a page the chat talks over the top of.

Acceptance

  1. Probes in internal/manual/chat_test.go in the asker's own words, including
    verbatim the one that failed:
    • can you use skills
    • do you have skills
    • what is a skill
    • why can't you use my skills
  2. A ## heading that answers the capability question directly, and states the
    refusal as well as the capability — per the manual law, "no, and here is what
    it says instead" is the most useful answer on the page. It has to name the
    switched-off case, because that is a state a person can be in without having
    chosen it knowingly.
  3. The probes fail before the heading exists and pass after. A probe added
    alongside the page that satisfies it proves nothing about retrieval.

Not in scope

Why the shelf is empty on this machine is a separate fault, being handled
elsewhere: memory.enabled off means no store, so discovery records nothing and
renderSkillCatalog emits zero bytes. Whether suppressing that section is right
when it leaves the model with no way to explain itself is also separate. This
issue is only that the corpus cannot answer the question, which is true
independently of both.

主要言語
Go
スター
115
フォーク
14
平均マージ
9時間 35分
マージ済み PR（30日）
752

環境構築

このプロジェクトの環境構築ファイルはまだ確認していません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

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。

Agent-Field/CodeAF のほかの issue

Agent-Field/CodeAF 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。