Release v10.2.0

MAKER Error Correction Framework (v10.2.0)

Integrates concepts from "Solving a Million-Step LLM Task with Zero Errors" (arXiv:2511.09030) to improve workflow reliability and scalability.

✨ New Features

Red-Flagging System: Discard suspicious agent outputs instead of repairing them

• Response length limits by operation type
• Format validation (strict mode - fail fast, don't repair)
• Uncertainty marker detection
• Circular reasoning detection
• Configuration: .spec-flow/config/red-flags.yaml

Multi-Agent Voting: First-to-ahead-by-k error correction for critical decisions

• Mathematical foundation: p_correct = p^k / (p^k + (1-p)^k)
• Temperature-based error decorrelation
• Voting-enabled operations: code review, breaking changes, security review
• New skill: .claude/skills/multi-agent-voting/SKILL.md
• Configuration: .spec-flow/config/voting.yaml

Task Complexity Scoring: 1-10 scale for task decomposition guidance

• Score 1-3 (Atomic): ~95% success rate
• Score 4-6 (Compound): ~85% success rate
• Score 7-10 (Complex): ~70% success rate → DECOMPOSE
• Automatic warnings for tasks scoring >5
• Updated: .claude/skills/task-breakdown-phase/reference.md

Error Rate Tracking: Historical success rates for adaptive model selection

• Per-task-type success rate tracking
• Cost per success (c/p) optimization guidance
• Model recommendations based on task complexity
• Configuration: .spec-flow/learnings/error-rates.yaml

📚 Documentation

• docs/maker-integration.md - Comprehensive MAKER concepts guide
• Updated CLAUDE.md with MAKER Error Correction section

---

Full Changelog: v10.1.0...v10.2.0