Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/tienenwu/fables/dispatching-subagentsnpx skills add tienenwu/fables --skill dispatching-subagentsgit clone --depth 1 https://github.com/tienenwu/fablesWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tienenwu/fables/dispatching-subagents)<a href="https://agentmods.dev/skills/tienenwu/fables/dispatching-subagents"><img src="https://agentmods.dev/badge/skills/tienenwu/fables/dispatching-subagents.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00082 | $0.02541 |
| Opus 5 | $0.00041 | $0.01270 |
| Sonnet 5 | $0.00016 | $0.00508 |
| Haiku 4.5 | $0.00008 | $0.00254 |
Grade A, and why
dispatching-subagents scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 74 lines — stays where its author put it; the contents beside it link to each section on GitHub.
🌐 English version · 繁體中文(正本 / canonical)
模型調度守則(dispatching-subagents)
目標:主對話 context 只裝結論與決策,執行細節外包。派工 prompt 模板見 references/templates.md;判斷測驗見 references/test-scenarios.md。
0. 什麼時候「不要」派工(先看這條,避免過度委派)
- 已知確切檔案與位置的單點查證(例:看某設定檔第 N 行的值)→ 自己 Read,比派 agent 便宜。
- 對話性回覆、解釋概念、改一兩行 → 自己做。
- 判準:已知確切檔案路徑或唯一關鍵字、且預計讀的內容 < 200 行 → 自己做;需要跨多目錄搜尋或猜命名 → 派工。
1. 指揮官不下場
以下工作一律派 subagent,主對話只收結論:
- 大量讀檔/掃 repo/「找出所有用到 X 的地方」→
Exploreagent(唯讀,最便宜) - 查網頁、讀文件、比較方案 →
general-purposeagent(可用 WebSearch/WebFetch) - 批次改檔(同一模式套用到多處)→
general-purposeagent - 設計實作方案 →
Planagent - 程式碼審查 →
code-revieweragent(另有system-architect做設計層第二輪)
1.5 平行派工(可平行就平行,序列是例外)
- 預設平行:2 個以上子任務彼此獨立(無共享狀態、無順序依賴),且唯讀或寫入範圍互不相交 → 同一則訊息一次派出,不要一個等一個。典型:多目錄搜尋、多維度審查(正確性/安全各派一個)、研究查證、read-back 驗收。
- 禁止平行:會寫同一批檔案(衝突),或後者輸入依賴前者結論(平行只是白跑)。寫入類要平行,先切出互斥的檔案範圍;切不開就序列。
- 規模上限:一批 ≤ 6 個。更大規模 fan-out(幾十個)屬重型多代理編排(
Workflow等,見 §3),維持「使用者明確要求才用」。 - 失敗隔離:批次中單一 agent 失敗不阻塞其他結果回收;失敗者單獨照 §5 升降級,不整批重派。
2. 派工三件套(每個派工 prompt 必含,模板見 references/templates.md)
- 目標與動機:要做什麼、為什麼(讓 agent 遇到岔路能自己判斷)。
- 驗收條件:可機械檢查的完成定義(例:「編譯通過並貼出 gradle 輸出最後 5 行」,不是「確保品質」)。
- 回報格式:只回結論 +
檔案:行號;長產物寫到指定路徑,回傳路徑。
3. model 與 effort 的實際可用值(2026-07-24 更新)
Agent工具的model參數:haiku/sonnet/opus/fable。省略 = 繼承主對話模型,這是預設正解。- 主迴圈預設跑 Opus(日常駕駛);
fable(Fable5)是最貴、消耗最快的層級,只保留給下面標「fable」的兩類,其餘一律不點。 Agent工具沒有 effort 參數;effort 由 agent 定義的 frontmatter 或 session 的 effortLevel 決定。Workflow工具(多代理編排)並非所有 harness 都有;只在使用者明確要求多代理編排、且工具確實存在時使用。- 選型基準(由便宜到貴):
haiku:機械性批次套用(模式已驗證過)、格式轉換、簡單 grep 彙整。- 省略(=主對話同級,預設 Opus):一般搜尋、實作、審查。
opus:難 bug 根因、非品味的高難度評審裁判(卡關升級鏈的終點,見 §5)。fable:只兩個用途——(1) 品味「判斷題」(UI 美感、文案語感的裁決);(2) 架構取捨/決策的裁決。丟足 context、拿一個裁決回來執行,不要讓它長時間留在迴圈裡動手。
- 動手打磨型品味(要跨多輪看 render 成品、逐輪迭代 UI)subagent 給不了——這時手動
/model把主迴圈切成 Fable5,收工切回 Opus。判準:「要 Fable5 看著成品逐輪改」→ 切主迴圈;「只要一個裁決/選一案」→ 派fablesubagent。 - 仍解不了的品味題/模糊題:向另一個模型家族要第二意見(
codex:rescue等)→ 給使用者選項並誠實標明信心低。
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 74 lines · 82 tokens per session scan A e8f5eee01a28
dispatching-subagents is a skill published in the GitHub repository tienenwu/fables (4 stars, last pushed 1mo ago), licensed MIT. It adds 82 tokens to every session and 2,541 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
speckit-specify
Create or update the feature specification from a natural language feature description.
speckit-analyze
Perform a non-destructive cross-artifact consistency and quality analysis across spec.md, plan.md, and tasks.md after task generation.
docs-build
Build, preview, and validate the Uno documentation website (DocFX) locally — surface real content errors vs expected noise, drive rendered pages with Playwright, and validate external-doc commit bumps in importexternaldocs.ps1 before a PR. Use when working under doc/, editing articles/ markdown, bumping an external…
speckit-taskstoissues
Convert existing tasks into actionable, dependency-ordered GitHub issues for the feature based on available design artifacts.
speckit-git-feature
Create a feature branch with sequential or timestamp numbering.
speckit-git-remote
Detect Git remote URL for GitHub integration.