2026-08-19 — Harness-neutral architecture

Reconstructed from RULES.md / tree at a25bcd3 (pre) vs 55ec8f4 (post).

Before

  • Forkable paste assumed Cursor-oriented paths in prose and skills.

  • Workstation facts: only $CODE_ROOT/MEMORIES.md + MEMORIES.example.md.

  • No HARNESS*, no general/harness*.md, no harness-setup skill.

After (55ec8f4)

  • New constant HARNESS: $CODE_ROOT/HARNESS.md.

  • New committed template HARNESS.example.md (variables: HARNESS_NAME, ALWAYS_ON_RULES, SKILLS_DISCOVERY_ROOT, LOCAL_RULES_DIR, …).

  • New general/harness.md, general/harness-boundary.md, .cursor/rules/harness-boundary.mdc.

  • New skill harness-setup; docs page harness-neutral.adoc.

  • Core rules: harness-neutral read of $HARNESS; template boundary (no machine paths in forkable files); chat file links via $HARNESS.

Naming note

Uppercase $CODE_ROOT/HARNESS.md was the first harness-neutral path; same-day rename moved to lowercase harness.md (see layer-rename detail).

Commit

  • 55ec8f4 — Make agent-rules harness-neutral with HARNESS.md machine layer