v0.6.0 — proactive frame-check guardrail

Proactive frame-check in scaffolded build folders

The deadliest failure mode is silently tuning a dead thesis for weeks while the AI keeps proposing the next calibration and never asks whether the premise is wrong. The existing "stay honest" guidance was passive text the build agent could forget. v0.6.0 makes it proactive.

Added

• Frame-check guardrail. /scaffold now also writes a Claude Code hook set (.claude/settings.json + .claude/hooks/*.sh) and a FRAME-LOG.md into the build folder:
  • a soft nudge before each git commit — is this tuning the thesis, or testing it?
  • a once-per-session Stop gate that blocks until the agent answers, re-reading the rejected alternatives if results disappoint,
  • a 4-field calibration record (altitude / model-rec / decision / prediction / outcome) logged to FRAME-LOG.md.
  • Portable POSIX sh (no jq), loop-safe, softenable; falls back to the passive CLAUDE.md guardrail outside Claude Code.
• /recheck reads FRAME-LOG.md — a long trail of tuning entries with no thesis-test is itself a calibration-myopia signal.
• Context template carries idea-level run-record fields so calibration data can accrue on the brainstorm side too.

Changed

• README banner re-rendered to the build-first arc (idea → brainstorm → build → recheck).

Install

• Claude CLI: /plugin marketplace add winchxyz/idea-to-build then /plugin install idea-to-build@idea-to-build
• Cowork / Claude Code: add that marketplace, or upload idea-to-build-plugin.zip below.