mirror of
https://github.com/walkinglabs/learn-harness-engineering.git
synced 2026-09-19 05:06:27 +08:00
Alec Chen
95a50dbdf7
The docs contradicted the checked-in project-01 code: the harness artifact list, the strip-the-app note, and the evidence-reset steps. The English page also lacked the run protocol, metrics, deliverables, and framing that zh-TW carried, and the zh-TW page imposed a 30-min/20-round limit and a git-branch workflow that lets a coding agent contaminate the weak-harness run. - en + zh-TW: mark the intro artifact list as a subset, add the docs/ links, the strip-the-app note, the data-dir symmetry rule, and reset-evidence guidance; align both pages to the same section flow (run protocol, how to measure, what to submit, framing, related lectures). - zh-TW: drop the 30-min/20-round limit and the git-branch workflow in favor of isolated run directories; quote the real starter/task-prompt.md; fix the init.sh description. - solution/CLAUDE.md: import @AGENTS.md so Claude Code loads the full harness spec (CLAUDE.md is auto-loaded, AGENTS.md is not). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>