v0.6.4

Rename the scenario-row behavioral-arc field script → guidance

"Script" invited the caller model to recite the field; it actually holds an arc to be pursued.
That mismatch degraded transcript realism — the evidence base every report finding rests on.

Clean cut, no alias. reps/rows.py requires guidance and rejects a legacy script key via a
dedicated ROW_LEGACY_RENAMED_INPUT_FIELDS, checked before the required-field loop so the error
names the rename rather than a confusing "missing guidance". Explicit rejection is required, not
cosmetic: input is an open-ended denylist, so a stale key would otherwise ride along silently.

Scope

• 116 JSONL rows across 12 banks — key only, 0 rows differ by anything else
• 10 drivers — token {scenario_input.guidance}; Behavior arc: label and canonical structure unchanged
• 5 pillar rubric judges — renamed, and no longer describe the caller as "scripted"
• 3 slot templates — {behavioral_arc_directives} → {behavioral_arc_guidance}
• Skills, _meta.md docs, reps/README.md, and the living 002 row contract
• Not migrated: tuned overlays under results/ (321 rows). They fail validation by design; the
  message tells operators to re-run reps-profile.

Also in this branch

README content gaps. Both READMEs advertised "eight skills and six slash commands" — there are
11 and 7, with all three reps-* skills missing from the table. Added the full inventory and
install instructions for Claude Code, Cursor, claude.ai, and the Claude API.

Full Changelog: v0.6.3...v0.6.4