Discord Github Correlator

Discord ↔ GitHub Issue Correlator

When to Use

Use this skill when:

• A user asks to "correlate", "match", or "find connections" between Discord discussions and GitHub issues
• They want to know which issues are being discussed in a Discord channel
• They need to prioritize issues based on community discussion activity
• They reference both a Discord server/channel and a GitHub repository

Trigger even when: The user doesn't explicitly say "correlation" — requests like "what GitHub issues relate to the auth-ig channel?" or "find issues discussed in Discord" should activate this skill.

Instructions

1. Gather Parameters

Required:

• Guild ID: Discord server ID (numeric string)
• Channel ID: Discord channel ID (numeric string)
• GitHub owner: Repository owner (e.g., anthropics)
• GitHub repo: Repository name (e.g., claude-code)

Optional:

• Keywords: Domain-specific keywords to look for (auth, performance, etc.)
• Message limit: How many recent Discord messages to fetch (default: 100)
• Top correlations: How many correlated issues to show (default: 15)

2. Fetch Discord Messages

Use the guildbridge MCP server:

import * as guildbridge from './servers/guildbridge';

// IMPORTANT: read_messages returns a BARE ARRAY of message objects
// ({ id, author, content, timestamp, ... }) — NOT an object with a
// `.messages` property. Newest-first; paginate older with `before`.
const messages: any[] = [];
let before: string | undefined;

while (messages.length < MESSAGE_LIMIT) {
  const batch = await guildbridge.read_messages({
    channel_id: CHANNEL_ID,
    limit: 50,
    before, // for pagination (id of the oldest message seen so far)
  });
  const arr = Array.isArray(batch) ? batch : batch?.messages ?? [];
  if (arr.length === 0) break;
  messages.push(...arr); // each item: { id, author, content, timestamp }
  before = arr[arr.length - 1].id;
}

Extract from each message:

• GitHub links (regex: https?://github\.com/[^\s)]+)
• Topics/keywords (case-insensitive matching)
• RFC references (regex: rfc\s*\d{4})

3. Fetch GitHub Issues

Use the github MCP server. Prefer search over list_issues for large repos:

import * as github from './servers/github';

// Approach A: Targeted search (recommended for keyword-heavy domains)
const searches = [
  `repo:${OWNER}/${REPO} is:open is:issue keyword1`,
  `repo:${OWNER}/${REPO} is:open is:issue keyword2`,
  // Add more searches for different keywords
];

for (const query of searches) {
  const response = await github.search_issues({ query, per_page: 100 });
  // Deduplicate by issue number
}

// Approach B: Full list (use only if search isn't applicable)
let cursor: string | undefined;
while (true) {
  const response = await github.list_issues({
    owner: OWNER,
    repo: REPO,
    state: 'open',
    first: 100,
    after: cursor
  });
  // ...
}

Extract: number, title, url, labels, createdAt, updatedAt, body (excerpt)

4. Score Correlations

For each issue, calculate a relevance score:

Topic matches (+2 per topic, weighted by Discord frequency):

• If a topic appears in the issue title/body, score += (Discord_mention_count × 2)

Direct links (+20 per link):

• If a Discord message links to this issue URL, strong signal

Label matches (+5 per relevant label):

• Check if issue labels include keywords (e.g., "auth", "mcp", "security")

Keyword overlap (+1 per significant keyword):

• Extract meaningful words from issue title (length > 4, not stopwords)
• If ≥2 keywords appear in a Discord message, score += matches

Recency boost (+5 if updated in last 30 days):

• Recent updates suggest active discussion

Comment activity (+0.2 per comment, max +10):

• High comment count indicates active discussion

5. Generate Report

Save three files to ./workspace/:

1. Raw Discord data: <channel-name>-messages.json
2. Raw GitHub data: <repo>-open-issues.json
3. Markdown report: <channel>-<repo>-correlation.md

Report structure:

# [Channel] ↔ [Repo] Correlation Report

Generated: [timestamp]

## Summary
- Discord messages analyzed: N
- Open issues analyzed: M
- Meaningful correlations found: K
- Top correlations shown: top 15

## Topics Extracted from Discord
| Topic | Mentions |
|-------|----------|
| topic1 | count |

## Top Correlated Issues
### 1. [#NUM] Issue Title
**Score:** X | **Comments:** Y | **Updated:** date
**URL:** link
**Labels:** label1, label2
**Matched Topics:** topic1, topic2
**Rationale:** matches topics X; directly linked in Y message(s); auth-related labels

<details>
<summary>Issue Description (excerpt)</summary>
> [First 300 chars of body]
</details>

---

## How to Read This Report
### Scoring
- Topic matches: weighted by frequency
- Direct links: +20 per link
- Auth labels: +5 per label
- Keyword overlap: multiple shared keywords

### Next Actions
1. Review top correlations
2. Cross-link Discord threads to issues
3. Triage using Discord context
4. Engage Discord participants in relevant issues

Gotchas

• read_messages return shape: it returns a bare array of message objects, not { messages: [...] }. Accessing response.messages yields undefined and silently looks like "0 messages" — the most common cause of an empty Discord result here. Always treat the result as the array itself (Array.isArray(batch) ? batch : batch?.messages ?? []).
• Discord pagination: read_messages with before returns older messages (pass the id of the oldest message seen). It is fast (≈1–3s for 50–80 messages).
• GitHub auth errors: list_issues may hit rate limits or auth failures mid-fetch. Wrap the loop in try-catch and save partial results.
• Pull requests in issue lists: Some APIs return PRs in issue endpoints. Filter with: if (issue.url?.includes('/pull/')) continue;
• Score normalization: Scores are relative, not absolute. A score of 40 vs 10 is meaningful; 40 vs 38 might not be.
• Large repos: anthropics/claude-code has 3500+ issues. Use search_issues with targeted queries instead of fetching all issues. Example: repo:anthropics/claude-code is:open is:issue authentication.
• Empty Discord channels: If a Discord channel is invite-only or you lack read permissions, the API returns 0 messages without error. Test with a known-readable channel first.

Example Usage

// For auth-ig Discord channel + anthropics/claude-code issues
const GUILD_ID = '1358869848138059966';  // MCP Contributors
const CHANNEL_ID = '1360835991749001368';  // auth-ig
const OWNER = 'anthropics';
const REPO = 'claude-code';

const AUTH_KEYWORDS = [
  'dpop', 'mtls', 'oauth', 'oidc', 'token refresh',
  'www-authenticate', 'authorization server', 'client registration',
  'scopes', 'bearer token', 'mcp auth'
];

// Follow steps 1-5 above

Related Tools

• guildbridge MCP server: read_messages, list_guilds, list_channels
• github MCP server: search_issues, list_issues, issue_read

Changelog

• 2026-06-21: Initial version based on auth-ig correlation task