跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
效率与协作

communications-triage-edho-ferdian

Channel-agnostic framework for triaging incoming messages (email, chat, Slack, LINE, Messenger, or any other channel) into four priority tiers, drafting replies that stay within a strict human-approval gate, and tracking whether a sent reply's promises actually get followed through on. Use when the user wants to build or apply a message-triage workflow, says "triase pesan", "atur inbox", "bantu balas email/chat", "klasifikasikan pesan masuk", "draft balasan", or asks how to keep track of promises made in a reply. Note: no channel (Gmail, Slack, etc.) is wired up yet in this ecosystem — this skill defines the triage logic and reply-drafting discipline to apply once a channel is connected, not a working integration.

edhoferdian · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add edhoferdian/EEF --skill communications-triage-edho-ferdian

技能说明


name: communications-triage-edho-ferdian description: >- Channel-agnostic framework for triaging incoming messages (email, chat, Slack, LINE, Messenger, or any other channel) into four priority tiers, drafting replies that stay within a strict human-approval gate, and tracking whether a sent reply's promises actually get followed through on. Use when the user wants to build or apply a message-triage workflow, says "triase pesan", "atur inbox", "bantu balas email/chat", "klasifikasikan pesan masuk", "draft balasan", or asks how to keep track of promises made in a reply. Note: no channel (Gmail, Slack, etc.) is wired up yet in this ecosystem — this skill defines the triage logic and reply-drafting discipline to apply once a channel is connected, not a working integration.

Communications Triage — Edho Ferdian Mode (Skill Edition) · v1.0

You are applying a message-triage discipline, not operating a mail client. This skill is channel-agnostic on purpose: at the time this was written, no channel integration (Gmail API, Slack MCP, LINE/Messenger bridge, etc.) is wired up anywhere in this ecosystem. Do not invent one. Do not pretend a tool call exists that doesn't. What you apply here is the classification logic, the reply-drafting discipline, and the follow-through tracking — the same logic a human would run by hand today, and the same logic a future channel integration would automate.

If the user has pasted message content directly into chat (an email body, a Slack thread, a screenshot's transcript), this skill applies fully to that pasted content — classify it, draft a reply, ask before anything resembling "send." What it does not do is reach out and fetch messages from a live inbox itself; nothing in this ecosystem is connected to one yet.

Why channel-agnostic, not "wire up Gmail now"

Building a specific integration ahead of an actual connected channel would mean maintaining fake tool calls, or code that talks to nothing. Per this ecosystem's ground-truth ethos (see dev-kickoff-edho-ferdian, spec-mining-edho-ferdian): never fabricate integrations that don't exist. This skill instead front-loads the part that survives any channel — the tiering rules, the drafting discipline, the follow-through contract — so that whenever a real channel is connected (an MCP server, a CLI, an API client), the only new work is the fetch/send plumbing. The triage brain does not need to be rewritten.

The four tiers

Every message gets classified into exactly one tier. Apply the checks in this order — the first match wins, don't keep evaluating after a hit:

1. skip — auto-archive, no summary needed

  • Sender is a no-reply / notification / alert address, or an automated system account (CI bots, ticketing-system bot comments, channel join/leave events, delivery receipts).
  • No human is waiting on a reaction to this message.
  • Action: note the count if triaging a batch ("12 pesan otomatis di-skip"), do not surface individual content.

2. info_only — summarize, no reply needed

  • The user is CC'd, not the primary recipient, and no question is directed at them.
  • Broadcast/announcement content (@channel/@here-style, newsletters, receipts, read-only file shares).
  • Group chat chatter where the user isn't named and isn't the last meaningful contributor.
  • Action: one-line summary per message (or per thread), nothing more.

3. meeting_info — cross-reference, no free-text reply needed

  • Contains a scheduling artifact: a video-call link, a date + meeting context, a room/location share, or a calendar-invite attachment.
  • Action: this tier is about reconciling, not drafting — check it against whatever calendar source is available (a connected calendar tool, or the user's own statement of their schedule) and flag conflicts or missing details. If no calendar source is connected, say so and surface the raw scheduling info instead of guessing availability.

4. action_required — draft a reply

  • A direct question awaiting an answer.
  • An explicit ask, scheduling request, or decision the user must make.
  • The user is @mentioned and a response is clearly expected.
  • Action: generate a draft reply (see below). Never send it — see the approval gate.

Ambiguous cases: when a message could plausibly fit two tiers, prefer the more attention-requiring tier (e.g., meeting_info over info_only if it's unclear whether a reply is expected) — treat under-triage as the worse failure than a message getting slightly more attention than it needed.

Draft-reply generation guidance

For every action_required message:

  1. Read the actual ask before drafting. Restate to yourself what specifically is being asked — a scheduling slot, a yes/no decision, a piece of information, an approval. A draft that doesn't answer the real question is worse than no draft.
  2. Match tone to the relationship, when that context is available. If the user maintains their own notes on how they talk to specific people or channels (a personal style file, prior message history, explicit instruction in the current conversation), use it. If no such context exists, default to a neutral, professional tone and say so — don't guess a familiarity level you have no evidence for.
  3. Keep commitments explicit and minimal. A draft that promises a specific date, a specific deliverable, or "I'll get back to you by Friday" is creating an obligation — see follow-through, below. Don't let a draft casually generate more promises than the user actually intends to keep.
  4. Flag uncertainty instead of hiding it. If the draft has to guess at something (e.g., availability, a fact the user hasn't confirmed), mark it inline — [VERIFY: ...] — rather than stating it as settled.
  5. Present the draft, don't act on it. See the approval gate below — this is non-negotiable regardless of how obviously c
阅读完整来源 ↗