/compress

description: Automatically compress CLAUDE.md or other memory files with protected-span safety. One-command token savings.
disable-model-invocation: false

Governor Compress

Automatically compress memory files with dense professional prose while strictly protecting critical spans.

Usage:

/governor:compress [light|medium|aggressive] [file]

Default target: `CLAUDE.md` Default level: `medium`

Compression targets:

• `light`: remove filler/repetition; target 15-30% savings outside protected spans.
• `medium`: collapse narrative into decision bullets; target 35-55% savings outside protected spans.
• `aggressive`: keep only rules, facts, commands, risks, decisions; target 50-70% savings outside protected spans.

One-command rule: do not ask the user to edit drafts, copy paths, or run follow-up commands unless manual mode is explicitly requested or a safety fallback is required.

Internal flow:

• Run Governor auto mode.
• Parse the returned JSON payload.
• Rewrite the protected payload internally.
• Run the JSON finalize command, then inspect `status`.
• If `status=quality_guard_failed` and `next_level` is present, retry once at

the stronger level using `retry_auto_command`.

• Finalize, validate, recover if needed, and restore the backup on unrecoverable

failure.

• Reject low-savings output with a quality guard.
• Return only the result summary.

Manual mode (`manual` or `--manual`) is only for very large files or explicit user request.

Protected spans are never modified:

• Code blocks and inline code.
• Paths, URLs, commands, env vars.
• API names, model names, versions, dates.
• Headings, frontmatter, tables.
• Warnings and irreversible rules.
• Brand and design tokens.

After compression, report:

• Original vs compressed token estimate.
• Memory saved percentage.
• Validation status.
• Quality-guard status.
• Backup location.