Discord Connector

Goal

Work with Discord conversations and channels in a way that is operationally useful, concise, and safe.

This skill is intended for:

• team coordination
• community management
• support triage
• incident communication
• internal handoffs based on Discord discussion

Workflow

1. Identify the user's intent: read, search, summarize, draft, post, or triage.

2. Identify the target Discord surface: server, channel, thread, message link, or user/role mention context.

3. Use available Discord tools (e.g., MCP Discord tools) to fetch only the necessary message history. Pass the narrowest parameters first — for example, target channel ID or thread ID, a small message limit (e.g., 50), and a time range if provided. Expand only if context is clearly missing.

   Example MCP tool invocation:

   tool: discord_get_messages
   params:
     channel_id: "112233445566778899"
     limit: 50
     before: "2024-06-01T00:00:00Z"   # omit if not filtering by time

   Adjust limit, before, and after incrementally rather than fetching large windows up front.

4. Validate the retrieved data before proceeding: confirm the message count is non-zero, the date range matches the user's intent, and no truncation warning is present. If any check fails, apply the relevant error-handling step below before continuing.

5. Normalize the conversation into signal: main topic, decisions made, unresolved questions, blockers, owner mentions, timestamps if relevant.

6. Produce the requested artifact in the format most useful for the task.

7. If posting back to Discord, keep the message short, readable, and channel-appropriate.

Error Handling

• No messages found: Inform the user that no messages were retrieved. Suggest narrowing the time window, checking channel permissions, or confirming the channel name.
• Permission denied / channel not accessible: Report the access failure clearly and ask the user to verify channel visibility or bot permissions before retrying.
• Channel or thread does not exist: Confirm the exact channel name or thread link with the user; do not guess or substitute an alternative.
• Incomplete history: If the retrieved window is clearly truncated, say so and note that conclusions may be partial before proceeding.

Operating Modes

1. Conversation Summary

Return: one-paragraph summary, key decisions, action items, unresolved questions.

2. Search And Retrieval

Return: what was found, who said it, where it was discussed, whether the result looks definitive or partial.

3. Announcement Drafting

Return: a polished Discord-ready draft, optional shorter alternative, optional call to action.

4. Support Or Incident Triage

Return: issue summary, impact, repro clues, urgency, missing information.

5. Handoff Extraction

Return: concise status, owner list, next steps, dependencies, risks.

Preferred Output Shapes

Choose the narrowest shape that matches the request.

Short Summary

## Summary

<2-4 sentence summary>

## Decisions

- <decision>

## Action Items

- <owner>: <task>

## Open Questions

- <question>

Search Result Brief

## What I Found

- <result>

## Sources

- <channel/thread/message context>

## Confidence

<high | medium | low>

Discord Post Draft

## Draft

<discord-ready message>

## Notes

- <optional context or tradeoff>

Rules

• Prefer exact channel and thread context over broad summary when the user asks about a specific discussion.
• Do not invent messages, users, timestamps, or decisions that are not supported by the retrieved conversation.
• Distinguish clearly between what was explicitly said and what is a reasonable inference.
• If the conversation is incomplete, say so directly.
• If posting back to Discord, optimize for readability: short paragraphs, light bullets, minimal jargon unless the channel is technical.
• Preserve sensitive boundaries: do not repost secrets, do not amplify private or irrelevant personal details, avoid quoting large blocks of chat unless the user explicitly asks.

Completion Checklist

Before finishing, verify:

• the response matches the user's actual Discord task
• the scope is limited to the relevant conversation
• any inferred content is labeled as inference
• the final output is usable without extra cleanup