Live survey304 models · 14 providers · 23,592 auditsAudited implement · SAFE →What changedHow we verifyllms.txt

Agent skill · Audited from source

caveman-cn

skill · juliusbrussee/cavemanApache-2.0 licenceAudited 13 Nov 2026 · source e01ff8e74d7e

★ 119,488pushed today
✕BLOCKGrade F · 30/100 · 25 findings · e01ff8e
119,488repo stars · whole juliusbrussee/caveman

Is it safe? No — BLOCK, grade F (30/100). We read this skill's source at commit e01ff8e74d7e and observed filesystem, network and shell access and a hard-coded secret (see the findings). The audit raised 25 findings, 11 critical or high. What it does: When to delegate to `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit) or `cavecrew-reviewer` (diff review) instead of working inline or using `Explore`. Their output is compressed, so main…

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “caveman-cn (juliusbrussee/caveman)”. Retrieved 13 Nov 2026. https://modelsatlas.com/skills/juliusbrussee/caveman-cn

Blocked. The page stays public so you can check this skill if someone recommended it to you. The audit found 11 critical or high findings, starting with “Inventory / provenance” in shared/platform/redact/payload.go. Do not install it without reading them.

2 skills are called “caveman-cn”. You are viewing juliusbrussee/caveman. 2 of them are BLOCK.

Trust audit

Audit 0.4.1 · source read 13 Nov 2026

Trust score30/100✕BLOCK F
0 critical11 high42 medium42 low1 info

BLOCK means the audit found something that could harm you or your machine. Grade F is shared by 129 of 18,389 audited skills.

What the source does

  • FilesystemDeclared7 observations
  • NetworkDeclared12 observations
  • ShellDeclared10 observations
  • DependenciesNot pinnedSome versions float
  • SecretsFoundSee the findings; values are never shown

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityFAIL6
L1 Static sourceSecrets, path traversal, supply chain, hidden codeFAIL18
L2 InstructionsPrompt injection, authority framing, hidden directivesFAIL1
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: no behavioural pass.

Findings (25)

HIGHInventory / provenanceinv.suspicious_name · shared/platform/redact/payload.go
F-13 Inventory / provenanceCWE-1104Layer L0
payload.go
Why
member named after an attack tool
Fix
remove or justify
HIGHCode injectioncode.eval_exec · browse/cdp.go:246
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
func (d *CDPDriver) Eval(ctx context.Context, expression string) (any, error) {
Why
evaluates text as code
Fix
remove; use a parser or a dispatch table
HIGHCode injectioncode.eval_exec · browse/cmd/caveman-browse/main.go:509
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
func (nilDriver) Eval(context.Context, string) (any, error) { return nil, context.Canceled }
Why
evaluates text as code
Fix
remove; use a parser or a dispatch table
HIGHCode injectioncode.eval_exec · browse/session.go:34
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
Eval(ctx context.Context, expression string) (any, error)
Why
evaluates text as code
Fix
remove; use a parser or a dispatch table
HIGHCode injectioncode.eval_exec · engine/ccr/store_sqlite.go:201
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
Exec(query string, args ...any) (sql.Result, error)
Why
evaluates text as code
Fix
remove; use a parser or a dispatch table
HIGHNetwork egressnet.metadata_ip · shared/platform/awscreds/awscreds.go:64
F-04 Network egressCWE-200 · CWE-319Layer L1
defaultIMDSBase      = "http://169.254.169.254"
Why
cloud metadata endpoint: the classic SSRF credential grab
Fix
—
HIGHNetwork egressnet.metadata_ip · shared/platform/awscreds/awscreds.go:467
F-04 Network egressCWE-200 · CWE-319Layer L1
netip.MustParseAddr("169.254.169.254"),
Why
cloud metadata endpoint: the classic SSRF credential grab
Fix
—
HIGHNetwork egressnet.metadata_ip · shared/platform/awscreds/awscreds.go:531
F-04 Network egressCWE-200 · CWE-319Layer L1
return fmt.Errorf("awscreds: refusing plaintext IMDS endpoint at host %q (allowed: loopback, 169.254.169.254, fd00:ec2::254)", u.Hostname())
Why
cloud metadata endpoint: the classic SSRF credential grab
Fix
—
HIGHNetwork egressnet.tls_off · shared/platform/postgresconfig/postgresconfig.go:74
F-04 Network egressCWE-200 · CWE-319Layer L1
config.ConnConfig.TLSConfig.InsecureSkipVerify = false
Why
certificate verification is disabled
Fix
leave verification on
HIGHNetwork egressnet.tls_off · shared/platform/postgresconfig/postgresconfig.go:81
F-04 Network egressCWE-200 · CWE-319Layer L1
fallback.TLSConfig.InsecureSkipVerify = false
Why
certificate verification is disabled
Fix
leave verification on
HIGHPrompt injectionprompt.credential_read · plugins/caveman/skills/caveman-compress/SKILL.md:13
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
> **Data leaves the machine.** Compression sends the file's full text to the model provider that `scripts/compress.py` is configured for: the `claude` CLI, the Anthropic API, or an OpenAI-compatible e
Why
asks the agent to read credentials
Fix
—
MEDIUMInventory / provenanceinv.binary · dist/caveman.skill
F-13 Inventory / provenanceCWE-1104Layer L0
caveman.skill
Why
a compiled or binary member cannot be reviewed from source
Fix
ship source, or explain the binary in the README
MEDIUMInventory / provenanceinv.binary · engine/pixel/assets/atlas-codepoints.bin.gz
F-13 Inventory / provenanceCWE-1104Layer L0
atlas-codepoints.bin.gz
Why
a compiled or binary member cannot be reviewed from source
Fix
ship source, or explain the binary in the README
MEDIUMInventory / provenanceinv.binary · engine/pixel/assets/atlas-offsets.bin.gz
F-13 Inventory / provenanceCWE-1104Layer L0
atlas-offsets.bin.gz
Why
a compiled or binary member cannot be reviewed from source
Fix
ship source, or explain the binary in the README
MEDIUMInventory / provenanceinv.binary · engine/pixel/assets/atlas-pixels.bin.gz
F-13 Inventory / provenanceCWE-1104Layer L0
atlas-pixels.bin.gz
Why
a compiled or binary member cannot be reviewed from source
Fix
ship source, or explain the binary in the README
MEDIUMInventory / provenanceinv.binary · engine/pixel/assets/atlas-wideflags.bin.gz
F-13 Inventory / provenanceCWE-1104Layer L0
atlas-wideflags.bin.gz
Why
a compiled or binary member cannot be reviewed from source
Fix
ship source, or explain the binary in the README
MEDIUMCode injectioncode.dynamic_import · packages/middleware/python/caveman_middleware/_httpx2.py:28
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
http = importlib.import_module(name)
Why
No further detail recorded for this rule.
Fix
—
MEDIUMCode injectioncode.dynamic_import · packages/middleware/python/caveman_middleware/llama_index.py:68
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
if isinstance(model, getattr(importlib.import_module(module), name)):
Why
No further detail recorded for this rule.
Fix
—
MEDIUMCode injectioncode.dynamic_import · packages/middleware/python/caveman_middleware/pydantic_ai.py:66
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
if isinstance(model, getattr(importlib.import_module(module), name)):
Why
No further detail recorded for this rule.
Fix
—
MEDIUMInsecure cryptocrypto.weak_random · evals/snapshots/reminder.json:28
F-09 Insecure cryptoCWE-327 · CWE-338Layer L1
"That's React's default behavior. When a parent re-renders, React re-renders every child it renders, whether or not the child's props changed. React doesn't compare props unless you ask it to.\n\n**To
Why
No further detail recorded for this rule.
Fix
—
MEDIUMNetwork egressnet.beacon_words · proxy/cmd/caveman-proxy/main.go:475
F-04 Network egressCWE-200 · CWE-319Layer L1
func withKeepalive(next http.Handler, beacon func()) http.Handler {
Why
No further detail recorded for this rule.
Fix
—
MEDIUMNetwork egressnet.beacon_words · proxy/cmd/caveman-proxy/main.go:478
F-04 Network egressCWE-200 · CWE-319Layer L1
beacon()
Why
No further detail recorded for this rule.
Fix
—
MEDIUMNetwork egressnet.raw_ip · .blocks/release-smoke.py:108
F-04 Network egressCWE-200 · CWE-319Layer L1
gateway = f"http://127.0.0.1:{port}"
Why
No further detail recorded for this rule.
Fix
—
MEDIUMObfuscation / stealthobf.base64_blob · engine/pixel/testdata/transform_anthropic/early_history_only.expected.json:17
F-07 Obfuscation / stealthCWE-506 · CWE-94Layer L1
"data": "iVBORw0KGgoAAAANSUhEUgAABiAAAALQCAIAAACWoC/QAAAxF0lEQVR4nO3dUW7kRrJAUS3C3/YevAZhtlRemPdXr0GCxchIMquolNohvXNhDHiSKTYlNQaDGLL8dpckSZIkSZImevuvb0CSJEmSJEnfOwMmSZIkSZIkTWXAJEmSJEmSpKl+34Dpj3/+fnr
Why
No further detail recorded for this rule.
Fix
—
MEDIUMObfuscation / stealthobf.base64_blob · engine/pixel/testdata/transform_anthropic/early_history_only.expected.json:25
F-07 Obfuscation / stealthCWE-506 · CWE-94Layer L1
"data": "iVBORw0KGgoAAAANSUhEUgAABiAAAADQCAIAAAABPz75AAAOmElEQVR4nO3dUXLjNhZAUS0i3+k9ZA2ubElZWPan6SKLIvAAgqKh7jx7zq354AFhmrJTXdVvRPXtIUmSJEmSJE10+69vQJIkSZIkSV87AyZJkiRJkiRNZcAkSZIkSZKkqQyYJEmSJEmSNJU
Why
No further detail recorded for this rule.
Fix
—

Audit 0.4.1 · source sha e01ff8e74d7e · 13 Nov 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • Publisherjuliusbrussee GitHub userOwner of the repository this skill is read from
  • Source folderjuliusbrussee/cavemanThe exact folder we audited
  • LicenceApache-2.0As the repository states it
  • Commit pinned for this verdicte01ff8e74d7eA new audit runs when the source changes
  • Repository statusNot archivedPushed today
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: When to delegate to `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit) or `cavecrew-reviewer` (diff review) instead of working inline or using `Explore`. Their output is compressed, so main…

---
name: cavecrew
description: >
  When to delegate to `cavecrew-investigator` (locate code), `cavecrew-builder`
  (1-2 file edit) or `cavecrew-reviewer` (diff review) instead of working inline
  or using `Explore`. Their output is compressed, so main context lasts longer.
---

Cavecrew = three subagent presets that answer in ultracave voice (tool runs: one status line in, one out; nothing between routine calls). Same job as Anthropic defaults (`Explore`, edit-style agents, reviewer); difference is the tool-result they return is compressed, so main context shrinks per delegation.

## When to use cavecrew vs alternatives

| Task | Use |
|---|---|
| "Where is X defined / what calls Y / list uses of Z" | `cavecrew-investigator` |
| Same but you also want suggestions/architecture commentary | `Explore` (vanilla) |
| Surgical edit, ≤2 files, scope obvious | `cavecrew-builder` |
| New feature / 3+ files / cross-cutting refactor | Main thread, or `feature-dev:code-architect` if installed |
| Review diff, branch, or file for bugs | `cavecrew-reviewer` |
| Deep code review with rationale + alternatives | `Code Reviewer` (vanilla) if installed, else main thread |
| One-line answer you already know | Main thread, no subagent |

Rule of thumb: **if you'd want the subagent's output in 1/3 the tokens, pick cavecrew. If you'd want prose, pick vanilla.**

## Why this exists (the real win)

Subagent tool results get injected into main context verbatim. A vanilla `Explore` that returns 2k tokens of prose costs 2k tokens of main-context budget every time. The same finding from `cavecrew-investigator` comes back as a compressed `path:line` table instead. Across 20 delegations in one session that's the difference between context exhaustion and finishing the task.

## Output contracts

What main thread can rely on per agent:

**`cavecrew-investigator`**
```
<Header>:
- path:line — `symbol` — short note
totals: <counts>.
```
Or `No match.` Always file-path-first, line-number-attached, backticked symbols. Safe to grep with `path:\d+`.

**`cavecrew-builder`**
```
<path:line-range> — <change ≤10 words>.
verified: <re-read OK | mismatch @ path:line>.
```
Or one of: `too-big.` / `needs-confirm.` / `ambiguous.` / `regressed.` (terminal first token).

**`cavecrew-reviewer`**
```
path:line: <emoji> <severity>: <problem>. <fix>.
totals: N🔴 N🟡 N🔵 N❓
```
Or `No issues.` Findings sorted file → line ascending.

## Chaining patterns

**Locate → fix → verify** (most common):
1. `cavecrew-investigator` returns site list.
2. Main thread picks 1-2 sites, hands paths to `cavecrew-builder`.
3. `cavecrew-reviewer` audits the diff.

**Parallel scout** (when investigation is broad):
Spawn 2-3 `cavecrew-investigator` calls in one message (different angles: defs vs callers vs tests). Aggregate in main thread.

**Single-shot edit** (when site is already known):
Skip investigator. Hand exact path:line to `cavecrew-builder` directly.

## What NOT to do

- Don't accept or reject a subagent's absence claim (`No match.`, `No issues.`, "section missing", "unchanged") without one targeted check of the relevant file, search scope, or diff. Check cited line numbers are within the file before acting.
- Don't use `cavecrew-builder` when you don't already know the file. Spawn investigator first or main thread will eat tokens passing context.
- Don't chain `cavecrew-investigator → cavecrew-builder` for a 5-file refactor. Builder will return `too-big.` and you'll have wasted a turn.
- Don't ask `cavecrew-reviewer` for "general feedback" — it returns findings only, no architecture opinions. Use `Code Reviewer` for that when installed.
- Don't expect prose. Cavecrew output is structured, sometimes terse to the point of cryptic. If a human will read it directly, paraphrase.

## Auto-clarity (inherited)

Subagents drop caveman → normal English for security warnings, irreversible-action confirmations, and any output where fragment ambiguity could be misread. Resume caveman after.

Excerpt, first 3,990 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

Install

Pinned to the audited commit e01ff8e

npx skills add JuliusBrussee/caveman -a kiro-cli -g
npx skills add JuliusBrussee/caveman -a codex --copy
git clone https://github.com/JuliusBrussee/caveman.git
npx skills add JuliusBrussee/caveman -g
npx skills add JuliusBrussee/caveman --skill caveman
npm install --ignore-scripts --no-audit --no-fund --no-package-lock --prefix packages/cli

Commands as the repository documents them: shown, never run by us. Our verdict covers commit e01ff8e74d7e only.

Works in which agents?

What its own text says, not a test

AgentStatusMore
Claude CodeMentioned in its text, not testedSkills for Claude Code · About Claude Code
Claude DesktopMentioned in its text, not testedSkills for Claude Desktop · About Claude Desktop
CodexMentioned in its text, not testedSkills for Codex · About Codex
GitHub CopilotMentioned in its text, not testedSkills for GitHub Copilot · About GitHub Copilot
CursorMentioned in its text, not testedSkills for Cursor · About Cursor
Gemini CLIMentioned in its text, not testedSkills for Gemini CLI · About Gemini CLI
OpenClawMentioned in its text, not testedSkills for OpenClaw · About OpenClaw
WindsurfMentioned in its text, not testedSkills for Windsurf · About Windsurf
“Mentioned” means the skill’s text names the agent. We have not run it there.

Audit history

Every audit, every source change

  1. ✕BLOCK F 30 / 100

    first audit · source e01ff8e74d7e

README badge

For maintainers · shows the latest audit

ModelsAtlas: BLOCK FMODELSATLASBLOCK · F
[![ModelsAtlas: BLOCK F](https://modelsatlas.com/skills/juliusbrussee/caveman-cn/badge.svg)](https://modelsatlas.com/skills/juliusbrussee/caveman-cn/)

The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.

Similar skills

Add this badge to your README

[![ModelsAtlas: BLOCK F](https://modelsatlas.com/skills/juliusbrussee/caveman-cn/badge.svg)](https://modelsatlas.com/skills/juliusbrussee/caveman-cn/)

Questions about caveman-cn

What does the caveman-cn skill do?
🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talking like a caveman.
Is caveman-cn safe to install?
No — not without reading the findings first. The audit graded it F (30/100) and found 11 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What can caveman-cn access on my machine?
The audit observed that it reaches the network, runs shell commands and reads or writes files. Each of those is consistent with what it says it does. Secrets in the source: found — see the findings.
Which assistants does caveman-cn work with?
Its documentation mentions claude-code, claude-desktop, codex, copilot, cursor, gemini-cli, openclaw and windsurf. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (e01ff8e74d7e), read on 2026-11-13. The repository is watched, and a new audit runs when it changes — this is the first audit.

Sources

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.