Changelog Orchestrator

Purpose

Orchestrate a 6-phase workflow to automate changelog generation: fetch data from multiple sources, synthesize narrative with AI, validate quality, and create pull requests.

Overview

This skill coordinates changelog automation through progressive disclosure:

• Phase 1: Initialize & fetch data from GitHub/Slack/Git
• Phase 2: AI synthesis (Writer Agent groups changes, generates narrative)
• Phase 3: Template formatting with frontmatter validation
• Phase 4: Quality review with two-layer gate (deterministic + editorial)
• Phase 5: PR creation with changelog file
• Phase 6: User handoff with summary and next steps

Integrates with MCP server (changelog_mcp.py) for deterministic operations while handling editorial work (synthesis, tone, quality judgment) directly.

Prerequisites

Environment:

• Python 3.10+ installed
• MCP server dependencies installed: pip install -r scripts/requirements.txt
• .changelog-config.json configured in project root

Required Tokens:

• GitHub: GITHUB_TOKEN environment variable with repo:read + repo:write scopes
• Slack (optional): SLACK_TOKEN environment variable with channels:history scope

Files:

• Config file: .changelog-config.json (see config/.changelog-config.example.json)
• Template: Markdown file with frontmatter (see assets/templates/default-changelog.md)

Instructions

Phase 1: Initialize & Fetch Data

1. Load Configuration

   • Call MCP tool get_changelog_config (no arguments for default .changelog-config.json)
   • If error: Display error message with suggestion (e.g., "Run /changelog-validate")
   • If success: Store config for later phases

2. Determine Date Range

   • For /changelog-weekly: Calculate start_date = today - 7 days, end_date = today
   • For /changelog-custom: Use provided start_date and end_date parameters
   • Validate dates are in ISO 8601 format (YYYY-MM-DD)
   • Ensure start_date < end_date

3. Fetch Data from All Sources

   • For each source in config["sources"]:
     • Call MCP tool fetch_changelog_data with:
       • source_type: From config (github/slack/git)
       • start_date: Calculated or provided date
       • end_date: Calculated or provided date
       • config: Source-specific config from file
     • Collect items from response data["items"]
   • Aggregate all items into unified dataset
   • Sort by timestamp (oldest to newest)

4. Display Fetch Summary

   • Total items fetched
   • Breakdown by source (e.g., "GitHub: 12 PRs, Slack: 5 messages")
   • Date range covered

Phase 2: Writer Agent - AI Synthesis

Role: You are now the Writer Agent. Your job is to transform raw data into user-friendly changelog content.

1. Group Items by Type

   • Features: Items with labels/types: enhancement, feature, new
   • Fixes: Items with labels/types: bug, fix, bugfix
   • Breaking Changes: Items with labels: breaking, breaking-change
   • Other: Everything else

2. Generate Narrative Summary

   • Write 2-3 sentences summarizing the overall theme of changes
   • Focus on what matters to users (not internal details)
   • Example: "This week focused on improving performance and user experience. Key highlights include a new dark mode toggle and faster page load times. Several critical bugs were fixed in the checkout flow."

3. Identify Top Highlights

   • Select 3-5 most impactful changes
   • Criteria: User-facing, significant value, breaking changes
   • Write 1-sentence description for each

4. Write Detailed Sections

   • For each type (features/fixes/breaking/other):
     • List items with title + link
     • Add brief context if needed
     • Use bullet points
   • Example:

     - Add dark mode toggle (#123) - Users can now switch between light and dark themes
     - Improve page load speed (#124) - 40% faster initial render
     

5. Apply Tone Guidelines

   • User-focused (not developer jargon)
   • Concise (no unnecessary details)
   • Professional (no slang or humor)
   • Active voice ("Added X" not "X was added")

6. Output: Draft markdown content (no frontmatter yet)

Phase 3: Formatter Agent - Template Compliance

Role: You are now the Formatter Agent. Your job is to apply the template structure and validate formatting.

1. Load Template

   • Read template file from config["template"]
   • Parse frontmatter placeholders (variables in {{...}})
   • Parse body structure (required sections)

2. Generate Frontmatter

   • Extract required fields:
     • date: End date from Phase 1 (YYYY-MM-DD)
     • version: Extract from latest git tag or increment (e.g., "1.2.0")
     • authors: Unique authors from all items (deduplicate)
     • categories: Derive from types (e.g., ["features", "fixes"])
   • Create frontmatter dictionary

3. Validate Frontmatter

   • Call MCP tool validate_frontmatter with:
     • frontmatter: Generated dictionary
     • schema_path: Optional (uses default schema)
   • If errors: Display and fix
   • If warnings: Note but continue

4. Apply Template Substitution

   • Replace all {{variable}} placeholders:
     • {{date}} → frontmatter.date
     • {{version}} → frontmatter.version
     • {{summary}} → From Phase 2
     • {{highlights}} → From Phase 2 (bullet list)
     • {{features}} → From Phase 2 (bullet list)
     • {{fixes}} → From Phase 2 (bullet list)
     • {{breaking_changes}} → From Phase 2 or "None" if empty
     • {{other_changes}} → From Phase 2 or "None" if empty
   • Combine frontmatter (YAML) + body (markdown)

5. Run Deterministic Quality Check

   • Call MCP tool validate_changelog_quality with:
     • content: Full markdown with frontmatter
   • Store quality score for Phase 4

6. Output: Complete formatted changelog markdown

Phase 4: Reviewer Agent - Quality Gate

Role: You are now the Reviewer Agent. Your job is to ensure the changelog meets quality standards before publishing.

1. Review Deterministic Score

   • Check MCP quality score from Phase 3
   • Required checks:
     • Frontmatter valid (YAML syntax)
     • All links accessible (no 404s)
     • Required sections present
     • Markdown syntax valid
   • Minimum score: 70/100 (deterministic only)

2. Editorial Review

   • Completeness: All fetched items represented (no missing changes)
   • Accuracy: No hallucinations (facts match source data)
   • Tone: User-friendly (no jargon, clear language)
   • Clarity: Easy to understand (no ambiguity)
   • Assign editorial score: 0-100

3. Calculate Combined Score

   • Combined = (Deterministic × 0.4) + (Editorial × 0.6)
   • Threshold: config["quality_threshold"] (default 80)

4. Decision

   • If combined score ≥ threshold: APPROVE → Proceed to Phase 5
   • If combined score < threshold: REJECT → Provide feedback

5. Feedback Loop (If Rejected)

   • Identify specific issues:
     • Missing changes: "Items X, Y not mentioned"
     • Tone issues: "Too technical in Features section"
     • Clarity issues: "Breaking changes unclear"
   • Return to Phase 2 with feedback
   • Maximum 2 iterations (if still fails after 2, warn user and allow override)

6. Output: Approval status + quality score + feedback (if any)

Phase 5: PR Writer Agent - Repository Operations

Role: You are now the PR Writer Agent. Your job is to commit the changelog and create a pull request.

1. Write Changelog File

   • Call MCP tool write_changelog with:
     • content: Approved markdown from Phase 4
     • output_path: From config["output_path"]
     • overwrite: true
   • Store file metadata (path, SHA256, line count)

2. Generate PR Description

   • Create markdown body:

     ## Summary
     Changelog for {start_date} to {end_date}
     
     ## Changes
     - Total: {count} items
     - Features: {feature_count}
     - Fixes: {fix_count}
     - Breaking: {breaking_count}
     
     ## Quality Score
     {combined_score}/100 (Deterministic: {det_score}, Editorial: {ed_score})
     
     ## Reviewer Checklist
     - [ ] All changes accurately represented
     - [ ] Tone is user-friendly
     - [ ] No sensitive information leaked
     - [ ] Links work correctly
     

3. Create Pull Request

   • Call MCP tool create_changelog_pr with:
     • branch_name: changelog-{end_date} (e.g., "changelog-2025-12-28")
     • commit_message: docs: add changelog for {end_date}
     • pr_title: Changelog for Week of {start_date} to {end_date}
     • pr_body: Generated description from step 2
     • base_branch: From config["pr_config"]["base_branch"] (default "main")
   • Store PR URL and number

4. Handle Errors

   • Branch exists: Suggest unique name or force delete
   • Permission denied: Check GitHub token has repo:write
   • Merge conflict: Suggest manual resolution

5. Output: PR metadata (URL, number, branch, diff summary)

Phase 6: User Handoff

Role: You are the Orchestrator Agent again. Present results to the user.

1. Display Success Summary

   🎉 Changelog PR created successfully!
   
   📋 Summary:
     - Date range: {start_date} to {end_date}
     - Changes: {count} items ({source breakdown})
     - Quality score: {score}/100
     - PR: {pr_url}
   
   Next steps:
     1. Review PR: Click link above
     2. Merge when ready: Changelog will be added to repo
     3. Iterate: Run /changelog-custom for different date ranges
   

2. Provide Next Actions

   • Review PR (link)
   • Merge when satisfied
   • Run /changelog-custom for different date range
   • Adjust quality threshold in config if needed

3. Save Reproducibility Bundle (Optional)

   • Create run_manifest.json with:
     • Execution timestamp
     • MCP server version
     • Data source versions
     • Item count per source
     • Quality scores
     • PR metadata
   • Store in project root or .claude/ directory

4. End Session

   • Thank user
   • Remind them of validation command: /changelog-validate

Output

Primary Artifact: GitHub Pull Request with changelog file

Secondary Artifacts:

• Changelog markdown file (e.g., CHANGELOG.md)
• Run manifest JSON (optional, for reproducibility)

Console Output: Phase-by-phase progress with status emojis

Error Handling

Error Type	Phase	Response	Recovery
Config file missing	Phase 1	Display error + suggest /changelog-validate	User creates config
GitHub token invalid	Phase 1	Display error + token setup instructions	User sets GITHUB_TOKEN
Template not found	Phase 3	Display error + template path	User fixes path in config
Quality gate fails	Phase 4	Provide feedback, iterate (max 2 rounds)	Lower threshold or fix content
PR creation fails (branch exists)	Phase 5	Suggest unique branch or delete existing	User resolves conflict

General Error Pattern:

1. Catch exception
2. Identify root cause
3. Display clear error message
4. Suggest specific fix
5. Provide fallback or recovery path

Examples

Example 1: Weekly Changelog (Happy Path)

User: /changelog-weekly

Phase 1: Fetched 20 items (GitHub: 12 PRs, Slack: 8 messages)

Phase 2: Generated summary + highlights + 4 sections

Phase 3: Applied template, validated frontmatter (✓)

Phase 4: Quality score: 92/100 (approved)

Phase 5: Created PR #1234

Phase 6: Displayed summary with PR link

Example 2: Custom Date Range

User: /changelog-custom start_date=2025-12-01 end_date=2025-12-15

Workflow: Same as Example 1, but uses provided date range

Example 3: Quality Gate Failure

Phase 4: Quality score: 68/100 (rejected, threshold 80)

Feedback: "Missing 3 PRs in Features section, tone too technical in Breaking Changes"

Phase 2 (iteration 2): Fixed issues

Phase 4 (retry): Quality score: 85/100 (approved)

Continue: Proceeds to Phase 5

Resources

Documentation:

• Architecture: 000-docs/000a-planned-plugins/changelog-automation/03-ARCHITECTURE.md
• User Journey: 000-docs/000a-planned-plugins/changelog-automation/04-USER-JOURNEY.md
• Technical Spec: 000-docs/000a-planned-plugins/changelog-automation/05-TECHNICAL-SPEC.md

MCP Server: Plugin-level script at scripts/changelog_mcp.py

Templates: Available in assets/templates/ directory

• default-changelog.md - Default format
• weekly-template.md - Weekly format (future)
• release-template.md - Release notes format (future)

Config Schema: config/.changelog-config.schema.json