guardex v7.0.0

• Breaking (soft). Consolidated 17 commands into 12 visible commands with flag-based subcommands. Five removed names (init, install, fix, scan, copy-prompt, copy-commands, print-agents-snippet, review) still work but print a one-line deprecation notice on stderr and will be removed in v8. See the migration table in "Copy-paste: common commands" above.

• Token-usage improvements. Trimmed the auto-installed agent templates that live inside every consumer repo and get loaded into every Claude/Codex session:

  • templates/AGENTS.multiagent-safety.md: 6990 B → 1615 B (−77%)
  • templates/codex/skills/guardex/SKILL.md: 2732 B → 1086 B (−60%)
  • templates/claude/commands/guardex.md: 472 B → 357 B (−24%)
  • Total: 10194 B → 3058 B per consumer repo (−70%, ~1.5k fewer tokens per agent session).

  The AI_SETUP_PROMPT and AI_SETUP_COMMANDS constants used by gx prompt are now compact checklists, so piping gx prompt into a model context is cheaper too.

• New gx prompt command replaces three prompt-emitting commands: gx prompt (full checklist), gx prompt --exec (commands only), gx prompt --snippet (AGENTS.md managed-block template).

• New flag surface on gx setup: --install-only (templates/hooks/locks only), --repair (fix drift), plus the existing --target, --parent-workspace-view, --dry-run, etc.

• New gx status --strict mirrors the old gx scan behavior (exit non-zero on findings).

• Updated internal REQUIRED_PACKAGE_SCRIPTS for consumer package.json so agent:safety:scan and agent:safety:fix helper scripts now invoke the new v7 surface (gx status --strict, gx setup --repair).