Release v1.6.0 — Uncapped Brainstorming + But-Why Integration · anotherben/claude-harness

What's New

Removed the 3-4 question hard cap. EXTRACT now uses a completeness-driven loop with 6 dimensions (Problem, User, Experience, Success, Boundaries, Tech Context), each requiring a "bedrock" answer before moving on.
But-why questioning techniques imported: one question per turn, build on answers (not sideways), contradiction detection, magic wand question for solution-first requests.
Respects user autonomy: "just go" stops EXTRACT immediately, remaining gaps marked as ASSUMED.
Scaling by tier updated: Micro (skip) → Small (1-3) → Medium (3-8) → Large (5-15+).

After producing the TDD, the agent self-challenges with 5 but-why questions before presenting to the user.
Catches weak designs, missing clarifications, and high-risk misunderstandings before they reach the user.

Before accepting a hypothesis, the agent must answer 4 challenge questions:
1. "Why is this the root cause and not a symptom?"
2. "Why would this only break now / in this case?"
3. "What would I expect if this hypothesis is WRONG?"
4. "Says who?" — evidence from code/data, not assumption?

Phase 3 now includes a successive "why" chain that traces root cause to bedrock (deliberate trade-off / nobody knew / legacy / copied code).
Bedrock interpretation rules guide the fix strategy.