ClaudeMods
☰
ZH-TW
● 0 人在線上 · 瀏覽 0 次
贊助提交作品
GitHub 儲存庫 · 發布者 alex2481kobe

agent-models

使用 agent.spawn 掛鉤,在原生代理清單的每一列標出子代理使用的模型和 effort。

alex2481kobe@alex2481kobe

alex2481kobe/claude-mods/tree/main/plugins/agent-models

原貼文圖片1
已翻譯

關於這個 mod

agent-models 是 Claude Code mods API 外掛,會在每個子代理啟動前把模型與 reasoning effort 標籤附加到任務描述,例如 Review auth · Opus 5.5 (high)。它從 Agent 工具的 model 選項、.claude/agents/ 與 ~/.claude/agents/ 中 agent 檔案的 model:/effort: frontmatter,或父代理模型解析標籤。Explore、Plan 等內建代理和外掛代理會保持不標記;未知模型不會猜測。安裝:/plugin marketplace add alex2481kobe/claude-mods;/plugin install agent-models@claude-mods。

安裝

請先查看作者 README,確認 marketplace 與外掛名稱;指令可能隨儲存庫結構而變動。

claude plugin marketplace add alex2481kobe/claude-mods
claude plugin install agent-models
原文 / README

agent-models

Each Claude subagent's model and effort in Claude Code's agent list.

Two Claude subagents in the agent list, labelled Opus 5.5 (high) and Haiku

Claude Code's agent list shows a subagent's task but not what it runs on. With this mod each row carries it, the way the codex mod's rows do:

Review auth · Opus 5.5 (high)
Check label.ts · Sonnet 5.5 (medium)
Say ok · Haiku

| The agent | Its label | | --------------------------------------------------- | ------------------------------------------------- | | general-purpose, a fork, an agent file's inherit | the parent's model and effort | | an agent file with model: (and effort:) | that model, and that effort | | the Agent tool's model option | that model; the parent's effort only if the same model | | Explore, Plan and other built-ins | none: their model is the engine's to choose | | a plugin's agent (codex:read) | none: the plugin labels its own |

Where the model cannot be known before the agent starts, the row is left as it was rather than guessed.

Requirements

  • Claude Code with mods (tested on 2.1.287)

Install

/plugin marketplace add alex2481kobe/claude-mods
/plugin install agent-models@claude-mods

How it works

  • An agent.spawn hook, which runs before the subagent starts, appends · <model> (<effort>) to the task's description; the agent list shows that description.
  • The model is the Agent tool's model option, else the agent file's, else the parent's for the agents that inherit it. Agent files are read from the project's .claude/agents/ and your own (~/.claude/agents/, or $CLAUDE_CONFIG_DIR/agents/), matched by their name.
  • The effort is the agent file's effort, else the effort of the parent's latest request, and that only when the agent runs on the parent's own model: another model may take no effort at all (Haiku takes none).
  • Model ids read as people say them: claude-opus-5-5 is Opus 5.5.

Limits

  • An agent file in a subfolder of agents/, or one defined in settings, a flag or policy, is not read, so its agent goes unlabelled.
  • An agent file with no model: line is left unlabelled.
  • An alias reads as its name without a version: haiku is Haiku.
  • An agent file's effort is shown even when the Agent tool's model option overrides its model.
  • Tried in the interactive terminal and headless (claude -p).
  • The mods API is early access and may change between Claude Code releases.

Develop

claude plugin validate plugins/agent-models
claude plugin test plugins/agent-models
claude --plugin-dir plugins/agent-models

其他同名作品

更多類似作品