Implement self-teaching Resume Issue Workflow

[VirtueMe]

Summary

• Transform issue docs(workflow): add standardized 'Resume Issue Workflow' template #58 from static documentation to self-teaching MCP workflow integration
• Embed workflow guidance directly into MCP server prompts and resources
• Enable automatic issue resumption without manual explanation

Changes

MCP Server Enhancements

• server.py - startup() - Enhanced with Resume Issue Workflow overview, trigger phrases, and standardized state keys
• server.py - get_session_context() - Added workflow object, best practices, and quick start guidance to session://context resource
• server.py - sign_off_prompt() - Enhanced with handoff state checklist for proper session transitions
• prompts.py - New module with STARTUP_PROMPT and SIGNOFF_PROMPT for future extensibility

Documentation Updates

• README.md - Added "Self-Teaching Workflows" section with Resume Issue Workflow explanation and example
• .claude/workflow-resume-issue.md - Human-readable reference (gitignored) explaining self-teaching concept, standardized state keys, and best practices

How It Works

Self-Teaching Concept:

• Traditional: Human writes docs → Claude might skip them → Inconsistent behavior
• New approach: MCP server shows prompts → Claude sees them every time → Consistent behavior

Workflow Trigger:
When user says "continue on issue #X", Claude automatically:

1. Signs on to claim session
2. Retrieves stored state from previous session
3. Presents findings to user
4. Asks for confirmation
5. Continues from exact point where previous session stopped

No manual explanation needed - the MCP server teaches Claude automatically.

Standardized State Keys

Defined standard keys for consistency across all sessions:

issue:{number}/status          → "in_progress"|"paused"|"complete"|"blocked"
issue:{number}/current_batch   → Batch number currently working on
issue:{number}/todos           → Array of remaining tasks
issue:{number}/next_steps      → Array of next actions for handoff
issue:{number}/worktree        → Path to worktree
issue:{number}/branch          → Branch name
issue:{number}/commits         → Array of commit hashes
issue:{number}/decisions       → Important decisions made
issue:{number}/blockers        → Blocking issues

Test Plan

Phase 2 testing (next step):

• Start work on an issue and store state
• Sign off and start new Claude session
• Say "continue on issue #X"
• Verify Claude retrieves state automatically
• Verify Claude presents findings and asks confirmation
• Verify Claude continues from saved state

Edge cases to test:

• Non-existent issue number
• No previous state (fresh start)
• Multiple active sessions
• Corrupted/missing state data

What Was Accomplished

Phase 1 (Complete):

• MCP server integration with self-teaching prompts
• Enhanced resources with workflow guidance
• Documentation for humans and AI
• Ready for testing

Phase 2 (Next):

• End-to-end testing with real scenarios
• Blog post ("The Claude Code Chronicles")
• Validation and refinement

Credit

Enhanced implementation by Cipher (DS-02) - transformed original static documentation approach into self-teaching MCP integration.

Success Criteria

✅ Workflow embedded in MCP server prompts
✅ Claude sees workflow guidance every time
✅ Standardized state storage keys defined
✅ Documentation for both humans and AI
✅ No manual explanation required

Next step: Test with real scenario to verify automatic workflow execution.

Closes #58

🤖 Generated with Claude Code