BumpyClock/agent-templates / skills_archive/caveman-compress
compress
VerifiedCompress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserve technical substance, code, URLs, and structure. Do compression directly with model judgment. Compressed version overwrites original file. Human-readable backup saved as FILE.original.md. Trigger: /caveman:compress <filepath> or "compress memory file"
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 4.5 KB |
| SECURITY.md | markdown | 2.0 KB |
| SKILL.md | skill md | 8.0 KB |
SECURITY.md
View on GitHub# Security
## Current model
`caveman-compress` is instruction-only now. Live behavior is defined in [`SKILL.md`](./SKILL.md), not by helper scripts.
Compression happens through normal agent behavior:
- filepath mode: read/write only the user-requested file plus sibling backup
- inline mode: compress user-provided text in memory and return it without file writes
No helper subprocess, no local validator loop, no custom CLI required.
## What the skill does
- In filepath mode, reads only file user explicitly points to
- Compresses prose directly with model judgment
- Preserves structure-heavy regions exactly
- In filepath mode, writes compressed result back to same path
- In filepath mode, saves sibling `.original.md` backup before overwrite
- In filepath mode, stops if backup already exists
- In inline mode, returns compressed text without creating files
## What the skill does NOT do
- Does not rely on helper scripts; live behavior is in `SKILL.md`
- Does not spawn subprocesses or shell commands for compression
- In filepath mode, does not scan outside requested path
- Does not silently overwrite existing backups
- Does not compress code/config files
- Does not rewrite code blocks, inline code, URLs, paths, commands, headings, table structure, prompt tags, or placeholders
## Why this is safer for complex files
Old scripted flow depended on brittle heuristics and validator regexes. That approach could damage prompt-heavy files with nested structure, XML-like tags, placeholder syntax, or mixed prose/code regions.
Current skill pushes preservation judgment into the model instructions themselves. When a region is ambiguous, the skill prefers leaving it alone over forcing more compression.
## Scope limits
- Natural-language files only
- Reject `*.original.md`
- Reject oversized files before rewrite
- Backup required before overwrite
- Inline mode returns text only; no file writes
## Reporting a vulnerability
If you believe you found a genuine security issue, open a GitHub issue with label `security`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |