Enrich message

Enrich Message

Enrich a draft message with code references, ticket links, and factual data. Outputs copy-pasteable markdown.

Preferences

Before starting, use the Read tool to read ~/.claude/skills/enrich-message/preferences.md. If the file does not exist, treat as "no preferences set" and use defaults.

Context

On startup, use Bash to detect: current git branch, remotes, repo root, and recent commits (last 5). Skip any that fail.

Command routing

Check $ARGUMENTS:

• help → display help then stop
• config → interactive setup then stop
• reset → delete preferences, confirm, stop
• anything else (including empty) → run the skill

Help

Enrich Message — Enriches draft messages with code references, tickets, and facts

Usage:
  /enrich-message                    Enrich draft from conversation context
  /enrich-message <url>              Enrich with a specific thread URL pre-loaded
  /enrich-message --brief            Keep enrichment concise
  /enrich-message --no-linear        Skip Linear ticket search
  /enrich-message config             Set preferences
  /enrich-message reset              Clear preferences
  /enrich-message help               This help

Examples:
  /enrich-message                    Paste your draft + context in chat, then run
  /enrich-message https://github.com/org/repo/pull/123#discussion_r456
                                     Pre-fetch this PR discussion thread

Current preferences:
  (from preferences.md or defaults)

Config

Use AskUserQuestion to collect:

• Q1: Default GitHub org — e.g., A-Teams-Network (used to resolve short PR refs)
• Q2: Writing style — Professional / Casual / Match original tone
• Q3: Reference depth — Minimal (links only) / Standard (links + brief context) / Deep (links + code snippets + explanation)

Save to ~/.claude/skills/enrich-message/preferences.md.

Reset

Delete ~/.claude/skills/enrich-message/preferences.md and confirm: "Preferences cleared. Using defaults."

First-time detection

If no preferences file exists, show:

First time using /enrich-message? Run /enrich-message config to set defaults, or just continue with sensible defaults.

Then proceed normally.

Workflow

Step 1: Parse inputs

Scan the current conversation for:

• Draft text — the user's draft response/comment (may be pasted as text or visible in a screenshot)
• Context URL(s) — GitHub PR comment links, Slack thread URLs, or any reference links
• Screenshots — read any images the user shared for additional context (comment threads, code diffs)
• Explicit instructions — any guidance like "add code references for X" or "mention the ticket for Y"

If $ARGUMENTS contains a URL, treat it as the primary context URL.

If inputs are ambiguous, ask:

I see your draft. What's the context? (paste a URL or describe the thread)

Step 2: Fetch thread context

Based on the context URL type:

GitHub PR comment (github.com/.../pull/N#discussion_rNNN):

• Extract owner, repo, PR number, and comment ID from the URL
• Run gh api repos/{owner}/{repo}/pulls/{pr}/comments to get the full comment thread
• Run gh pr view {pr} --repo {owner}/{repo} --json title,body,files,headRefName for PR context
• Identify the specific file and lines under discussion

GitHub PR review (github.com/.../pull/N#pullrequestreview-NNN):

• Run gh api repos/{owner}/{repo}/pulls/{pr}/reviews/{review_id}/comments for review comments
• Run gh pr view for PR context

GitHub issue (github.com/.../issues/N):

• Run gh issue view {n} --repo {owner}/{repo} --json title,body,comments

Slack thread (if URL provided):

• Note: Slack MCP not available — ask the user to paste the thread content

No URL / screenshot only:

• Extract context from the screenshot content
• Use visible file paths, function names, and discussion text as search seeds

Step 3: Analyze the topic

From the draft and context, identify:

• Key claims — what technical assertions does the draft make? (e.g., "we concatenate files on frontend")
• Technical concepts — function names, component names, patterns, architecture decisions mentioned
• Missing references — where the draft says something factual but doesn't link to evidence
• Search seeds — keywords, file paths, function names to search for in the codebase

Step 4: Search codebase

For each search seed identified:

1. Use Grep to find relevant code across the repo (function definitions, variable names, patterns)
2. Use Read to examine matched files for surrounding context
3. Note file paths and line numbers for each reference found

Build code references as GitHub permalink URLs:

• Format: https://github.com/{owner}/{repo}/blob/{branch}/{path}#L{line}
• Use the current branch or main as appropriate
• If the code is from a PR branch, use that branch's ref

Step 5: Search Linear (unless --no-linear)

Search Linear for tickets related to the discussion topic:

1. Use mcp__claude_ai_Linear__list_issues with keywords from the topic
2. For relevant matches, get details with mcp__claude_ai_Linear__get_issue
3. Note ticket identifiers (e.g., AIS-123) and URLs for linking

Step 6: Search GitHub for related work

Look for additional context:

1. gh pr list --repo {owner}/{repo} --search "{keywords}" --state all --limit 5 — find related PRs
2. gh search commits "{keywords}" --repo {owner}/{repo} --limit 5 — find related commits
3. Check other branches if the discussion mentions work-in-progress or upcoming changes

Step 7: Compose enriched message

Rewrite the draft with enrichments:

1. Preserve the author's intent — keep the core message and tone intact
2. Add inline code references — link to specific files/lines: [filename.ts#L42](permalink)
3. Add ticket references — mention Linear tickets: [AIS-123](url) where relevant
4. Add PR/commit references — link related PRs: [PR #123](url) or commits
5. Improve clarity — fix grammar, tighten phrasing, but don't over-formalize unless configured
6. Format for the platform:
   • GitHub comments: use GitHub-flavored markdown, <details> blocks for long code snippets
   • Slack: use simpler markdown (bold, links, code blocks)
   • Default: GitHub markdown

Writing style rules:

• Match the formality of the original draft unless preferences say otherwise
• Keep it concise — --brief means links only, no extra explanation
• Every factual claim should have a reference (code link, ticket, PR)
• Use relative references where natural ("as implemented in meetings-tab.tsx") with the link on the filename

Step 8: Present output

Always copy to clipboard — write the enriched message to /tmp/enriched-message.md using the Write tool, then run pbcopy < /tmp/enriched-message.md via Bash. This ensures the user can directly Cmd+V without losing markdown formatting.

After copying, confirm briefly:

Copied to clipboard. {N} code references, {N} Linear tickets, {N} PR links.

Sources: {list of repos/branches checked}

If any references couldn't be found, mention:

Could not find code references for: {list}

Principles

• Factual, not fabricated — only reference code, tickets, and links that actually exist. Never invent file paths or line numbers. If a reference can't be found, say so.
• Preserve the author's voice — enrich, don't rewrite. The output should sound like the user, just better-sourced.
• Links must be valid — every [text](url) must point to a real, accessible URL. Use GitHub permalinks with actual commit SHAs or branch names.
• Read-only — never post, comment, or modify anything. The user decides when and where to paste.
• Platform-aware formatting — GitHub comments support full GFM; Slack needs simpler markdown. Format accordingly.