Skills caveman-compress

📦

caveman-compress

Name: caveman-compress
Author: juliusbrussee

Content revision r1 High Risk 📁 Filesystem access⚙️ External commands🔑 Env variables

Compress Project Memory Files

Large project memory files consume tokens every time an AI coding session starts. This skill compresses prose memory files while preserving technical details and saving a readable backup.

Supports: Claude Codex Code(CC)

⚠️ 38 Poor

Stop for confirmation before installing.

Review the plan and obtain explicit user consent before changing files.

Install with my Agent

Copy this request to your Agent. It includes the canonical Skill page and manifest.

Agent request

Review the Skillstore skill "caveman-compress" from https://skillstore.io/skills/juliusbrussee-caveman-compress.md and its manifest at https://skillstore.io/api/skills/juliusbrussee-caveman-compress/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

Your Agent should still show its plan and request any confirmation required by the security policy.

Agent-readable resources

Use these links when an AI agent, crawler, or script needs clean context instead of reading the full page.

Markdown detail GET /skills/juliusbrussee-caveman-compress.md Signed manifest GET /api/skills/juliusbrussee-caveman-compress/manifest Signed lockfile GET /api/skills/juliusbrussee-caveman-compress/lockfile

Test it

Using "caveman-compress". Compress CLAUDE.md.

Expected outcome:

Compression completed. CLAUDE.md now contains compressed instructions, and CLAUDE.original.md contains the readable backup.

Using "caveman-compress". Compress project-notes.md with validation.

Expected outcome:

Compressed file saved after validation. Headings, links, code blocks, and paths were preserved.

Using "caveman-compress". Try to compress a configuration file.

Expected outcome:

The file was skipped because it was classified as code or configuration, not natural language prose.

Security Audit

High Risk

v6 • 7/9/2026 Open versioned report

Most static findings are false positives from markdown backticks, regex patterns, comments, and documented safe credential use. The meaningful residual risks are that selected files are sent to Claude and model output can overwrite future agent memory files.

Files scanned

964

Lines analyzed

Review items

False positives ignored

Confirmed security concerns (2)

High

scripts/compress.py:65-80 scripts/compress.py:145-147

Persistent Prompt Injection Through Compressed Memory

The model output is written back to the original memory file after compressing untrusted file text. A hostile source file could influence the model to add future agent instructions to CLAUDE.md.

The prompt includes the original content and the resulting text overwrites the memory file. Exploitation requires malicious content in the selected file, so confidence is high but not absolute.

Medium

scripts/compress.py:43-47 scripts/compress.py:79-80 scripts/compress.py:141-143

External LLM Submission of Local File Content

The selected file content is embedded in a Claude prompt and sent through the Anthropic SDK or Claude CLI. Users can disclose sensitive memory notes if they compress confidential files.

The data flow is direct: original file text is inserted into the prompt and submitted to Claude. The risk depends on user-selected file sensitivity, but the external submission is clear.

Risk Factors

📁 Filesystem access (2)

scripts/cli.py:48 scripts/validate.py:11

⚙️ External commands (15)

scripts/compress.py:53 scripts/compress.py:16-21 scripts/compress.py:21 scripts/compress.py:70-75 scripts/validate.py:6-43 SECURITY.md:9 SKILL.md:14 SKILL.md:18 SKILL.md:22 SKILL.md:49-50 SKILL.md:50-52 SKILL.md:52 SKILL.md:53 SKILL.md:57 SKILL.md:74

🔑 Env variables (7)

scripts/compress.py:37 scripts/compress.py:44 scripts/compress.py:37 scripts/compress.py:38 scripts/compress.py:42 SECURITY.md:9 SECURITY.md:23

Audited by: codex View Audit History →

Share & cite this report

Share the versioned assessment report, neutral badge, embed card, and citations. Skillstore reports evidence without deciding whether this Skill is safe.

Open versioned report

Copy report link

https://skillstore.io/skills/juliusbrussee-caveman-compress/audits/6?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown badge

[![Skillstore security assessment](https://skillstore.io/badges/skills/juliusbrussee-caveman-compress/security.svg)](https://skillstore.io/skills/juliusbrussee-caveman-compress?utm_source=security_passport_badge)

HTML badge

<a href="https://skillstore.io/skills/juliusbrussee-caveman-compress?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/juliusbrussee-caveman-compress/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Embed card

<iframe src="https://skillstore.io/embed/skills/juliusbrussee-caveman-compress.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Academic citations (APA · BibTeX · CFF)

APA citation

juliusbrussee. (2026). caveman-compress security audit report (audit version 6) [Author version unspecified]. Skillstore. https://skillstore.io/skills/juliusbrussee-caveman-compress/audits/6

BibTeX citation

@techreport{juliusbrussee-juliusbrussee-caveman-compress-2026,
  author = {juliusbrussee},
  title = {caveman-compress security audit report (audit version 6)},
  institution = {Skillstore},
  year = {2026},
  number = {6},
  url = {https://skillstore.io/skills/juliusbrussee-caveman-compress/audits/6},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "caveman-compress security audit report (audit version 6)"
version: "unspecified"
type: report
authors:
  - name: "juliusbrussee"
date-released: "2026-07-09"
url: "https://skillstore.io/skills/juliusbrussee-caveman-compress/audits/6"
identifiers:
  - type: other
    value: "skillstore:juliusbrussee-caveman-compress:audit:6"
    description: "Skillstore immutable audit report identifier"

Skillstore Score

Why this score Evidence Confidence: High

Architecture

Maintainability

Content

Community

Spec Compliance

What You Can Build

Reduce Recurring Project Context

Compress CLAUDE.md so each Claude Code or Codex session loads fewer repeated tokens.

Maintain Compact Team Notes

Shorten project preferences, todo files, and onboarding notes while keeping links and commands intact.

Prepare Memory Files for Long Projects

Trim large prose memory files before starting long-running tasks that need stable context.

Try These Prompts

Compress One Memory File

Use caveman-compress on CLAUDE.md and tell me where the backup was saved.

Compress Project Notes

Compress docs/project-notes.md, preserve headings and links, and report any validation warnings.

Check Before Compression

Check whether todos.md is a safe natural language target, then compress it only if it passes detection.

Plan Memory Cleanup

Find project memory files that are safe compression candidates, explain the order, and compress selected files one at a time with validation.

Best Practices

Review the readable backup before committing compressed memory changes.
Remove secrets and private notes before sending a file for compression.
Use the skill on focused memory files instead of broad mixed-content documents.

Avoid

Do not use it to compress source code, environment files, or configuration files.
Do not ignore validation warnings when paths, links, or bullets changed unexpectedly.
Do not run it on untrusted memory files without reviewing the output first.