Speckit + Claude Code Skills = fly by wire

[Neno73]

💡 Spec-Kit Autopilot: Making SDD Conversational and Self-Correcting

speckit-autopilot.zip
The Problem
When using spec-kit with AI agents, changes require manual intervention:

User: "Actually, let's use React instead of Blazor"
Result: User must manually update plan.md, tasks.md, research.md, then re-invoke commands
Pain: Cognitive overhead to remember what needs updating, easy to create inconsistencies

The Idea
Create a Claude Skill that watches conversations and automatically orchestrates spec-kit commands when it detects pivots, changes, or ambiguities.
Example workflow:
User: "Switch to Go instead of Node.js"

🎯 Autopilot: Detected tech stack pivot
→ Invokes /speckit.clarify for Go ecosystem choices
→ User answers: "Gin, Postgres, REST"
→ Invokes /speckit.plan with full context
✅ All artifacts updated, ready to continue
How It Works
The skill monitors for trigger patterns:

Tech pivots: "switch to", "use X instead", "migrate to"
Requirements: "add feature", "remove", "also include"
Principles: "we should always", "add rule"
Agent ambiguity: When implementation hits unclear specs

Then automatically:

Asks minimal clarifying questions (or uses /speckit.clarify)
Invokes the right spec-kit command with full context
Keeps all artifacts synchronized

Benefits
✅ Fly-by-wire development - Focus on intent, not file management
✅ No manual artifact updates - Spec-kit commands invoked automatically
✅ Self-correcting - Catches ambiguities before they become bugs
✅ Leverages existing spec-kit - No spec-kit changes needed
✅ Optional - Install as a Claude Skill, doesn't affect core workflow
Implementation
Built as a Claude Agent Skill:

Single SKILL.md file (~300 lines)
No dependencies or scripts
Uses spec-kit's existing commands
Works with Claude Code, Claude.ai, and API

Status
✅ Prototype complete - Fully functional skill ready for testing, see attached
speckit-autopilot.zip

📦 Packaged - Can be installed immediately
🧪 Needs validation - Looking for community feedback and real-world testing
Questions for Community

What other trigger patterns should activate the autopilot?
How intrusive should clarifications be? (Batch vs. immediate)
Should this be integrated into spec-kit core or remain a separate skill?
What else could benefit from this "intelligent orchestration" approach?

Would love feedback! Is this solving a real pain point you experience? What would make it more useful?

[samikh2020]

i like the idea - have you been using it since? and whats your experience been like?