Skip to main content

action-suggester

Generate non-binding follow-up action suggestions from lead summaries or lead lists. Use when users ask for next best actions, call list for hot leads, or follow-up draft plan without automatic execution. Recommended chain: summary-generator then action-suggester then supervisor approval. Do not use for parsing, storage, or autonomous action execution.

来源信息

仓库
modbender/skill-library-mcp
最近来源活动
2026年3月6日 18:33
检测到的 SKILL.md 语言
英语
星标
12
分支
3

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
action-suggester
description
Generate non-binding follow-up action suggestions from lead summaries or lead lists. Use when users ask for next best actions, call list for hot leads, or follow-up draft plan without automatic execution. Recommended chain: summary-generator then action-suggester then supervisor approval. Do not use for parsing, storage, or autonomous action execution.
# Action Suggester Produce ranked follow-up suggestions without taking any external action. ## Quick Triggers - Suggest next actions for today's P1 leads. - Build a call/email/visit plan from summary. - Draft follow-up queue for unresolved high-priority leads. ## Recommended Chain `summary-generator -> action-suggester -> supervisor confirmation` ## Execute Workflow 1. Accept input from Supervisor. 2. Validate input with `references/action-input.schema.json`. 3. Apply deterministic prioritization rules to propose follow-up items. 4. Emit actions with: - `action_type` in `call`, `email`, or `visit` - `lead_id` - `description` 5. Validate output with `references/action-output.schema.json`. 6. Return suggestions for human or Supervisor review only. ## Enforce Boundaries - Never execute suggested actions. - Never send messages, emails, or calendar invites. - Never write to database or stateful systems. - Never parse raw chat exports. - Never approve its own output for execution. ## Handle Errors 1. Return an empty action list when evidence is insufficient. 2. Reject inputs that fail schema validation. 3. Surface deterministic rule conflicts in plain text diagnostics.
在 GitHub 查看