📦

dcg

Name: dcg
Author: Dicklesworthstone

Content revision r1 Critical ⚙️ External commands🌐 Network access📁 Filesystem access

Block Destructive Claude Code Commands

AI coding agents can accidentally run commands that delete work or damage infrastructure. This skill adds a fast guardrail for Claude Code Bash hooks that blocks known destructive patterns before execution.

Supports: Claude Codex Code(CC)

⚠️ 38 Poor

Do not auto-install this skill.

The canonical policy requires operator review before any installation action.

Install with my Agent

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

Agent request

Review the Skillstore skill "dcg" from https://skillstore.io/skills/dicklesworthstone-dcg.md and its manifest at https://skillstore.io/api/skills/dicklesworthstone-dcg/manifest. Verify the artifact. Do not auto-install. Inspect the skill and report your findings, then wait for an operator or manual installation decision.

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/dicklesworthstone-dcg.md Signed manifest GET /api/skills/dicklesworthstone-dcg/manifest Signed lockfile GET /api/skills/dicklesworthstone-dcg/lockfile

Test it

Using "dcg". Test whether git reset --hard should run through Claude Code.

Expected outcome:

The guard blocks the command and explains that it destroys uncommitted changes. It suggests saving work first before running the command manually.

Using "dcg". Configure only Docker and Kubernetes protections.

Expected outcome:

Enable the containers.docker pack.
Enable the kubernetes pack or selected sub-packs.
Restart Claude Code after updating the hook configuration.

Using "dcg". Check whether rm -rf in a temp directory is allowed.

Expected outcome:

The documented policy allows recursive deletion inside temporary directories, but blocks similar commands against home or project paths.

Security Audit

Critical

v9 • 7/9/2026 Open versioned report

Most static findings are false positives caused by documentation of commands that DCG is designed to block. The real risks are the quick-install curl-to-bash patterns, the sudo installer path, persistent Claude Code hook configuration, and the documented bypass variable.

Files scanned

455

Lines analyzed

Review items

False positives ignored

Confirmed security concerns (4)

Critical

SKILL.md:197

Pipe to shell pattern

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The finding points to a curl-to-shell installer pattern that executes remote content. This is a real installation risk.

Critical

SKILL.md:200

Pipe to shell pattern

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The finding points to a curl-to-shell installer pattern that executes remote content. This is a real installation risk.

Critical

SKILL.md:197-203

Remote installer piped directly to shell

The installation section recommends curl commands that fetch an install script from GitHub and pipe it directly into bash, including a sudo variant. This creates a supply-chain risk if users run the commands without reviewing the downloaded script.

The README explicitly shows curl -fsSL piped into bash and sudo bash in the installation section. This is user-facing execution guidance, not only a blocked-command example.

Medium

SKILL.md:190 SKILL.md:426

Documented bypass weakens guardrail

The skill documents DCG_BYPASS=1 as an escape hatch and troubleshooting workaround. This is transparent, but teams should treat it as a control bypass in shared agent environments.

The bypass variable is explicitly documented in the environment variable table and troubleshooting section. It is legitimate but materially affects the security control.

Capability review items (14)

These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.

Critical

SKILL.md:197

Shell command substitution

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The installer URL embeds command substitution and pipes downloaded content directly to bash. This is a real supply-chain and remote-code-execution risk for users who follow the instructions.

Critical

SKILL.md:200

Shell command substitution

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The easy-mode installer repeats the remote curl-to-bash pattern with command substitution. It can execute changed remote content on the user machine.

Critical

SKILL.md:203

Shell command substitution

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The system-wide installer pipes remote content into sudo bash. This grants downloaded code elevated privileges.

Critical

SKILL.md:203

sudo privilege escalation

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

Line 203 pipes a remote install script into sudo bash for system-wide installation. This is privileged remote code execution if a user follows it.

Critical

SKILL.md:203

Hardcoded URL

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The hardcoded GitHub URL is used for a sudo remote installer. This combines network fetch with elevated execution.

High

SKILL.md:196-204

Ruby/shell backtick execution

```bash

The code block contains curl-to-shell installation commands. Even as documentation, this encourages executing remote code without review.

High

SKILL.md:196-204

Template literal with command substitution

```bash

The fenced install block includes command substitution in remote installer commands. The surrounding context is user-facing installation guidance, not inert analysis.

High

SKILL.md:197

Hardcoded URL

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The hardcoded GitHub URL is used for a remote installer that is piped to bash. The network dependency controls executed installation content.

High

SKILL.md:200

Hardcoded URL

curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/destructive_command_guard/master/ins

The hardcoded GitHub URL is used for the easy-mode remote installer. Users execute content fetched from that URL directly.

High

SKILL.md:218

Hidden file in home directory

Add to `~/.claude/settings.json`:

The instructions modify ~/.claude/settings.json to add a persistent PreToolUse hook. This is expected behavior but affects a hidden user configuration file.

High

SKILL.md:418

Hidden file in home directory

1. Verify `~/.claude/settings.json` has hook configuration

The troubleshooting step depends on ~/.claude/settings.json hook configuration. It is legitimate but confirms reliance on hidden user configuration.

Medium

SKILL.md:209

Hardcoded URL

cargo +nightly install --git https://github.com/Dicklesworthstone/destructive_command_guard

The cargo install command fetches and builds from a GitHub repository. This is legitimate installation guidance but still executes externally sourced code.

Medium

SKILL.md:218

Hidden file access

Add to `~/.claude/settings.json`:

The skill asks users to add hook configuration under ~/.claude/settings.json. Persistent hook changes should be reviewed before installation.

Medium

SKILL.md:418

Hidden file access

1. Verify `~/.claude/settings.json` has hook configuration

The troubleshooting guidance references hidden Claude settings. This is expected for a hook but remains sensitive configuration access.

Risk Factors

⚙️ External commands (50)

🌐 Network access (4)

SKILL.md:197 SKILL.md:200 SKILL.md:203 SKILL.md:209

📁 Filesystem access (7)

SKILL.md:173 SKILL.md:218 SKILL.md:418 SKILL.md:173 SKILL.md:218 SKILL.md:418 SKILL.md:92

Detected Patterns

Pipe to shell pattern×2

Audited by: claude 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/dicklesworthstone-dcg/audits/9?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown badge

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

HTML badge

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

Embed card

<iframe src="https://skillstore.io/embed/skills/dicklesworthstone-dcg.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

Dicklesworthstone. (2026). dcg security audit report (audit version 9) [Author version unspecified]. Skillstore. https://skillstore.io/skills/dicklesworthstone-dcg/audits/9

BibTeX citation

@techreport{dicklesworthstone-dicklesworthstone-dcg-2026,
  author = {Dicklesworthstone},
  title = {dcg security audit report (audit version 9)},
  institution = {Skillstore},
  year = {2026},
  number = {9},
  url = {https://skillstore.io/skills/dicklesworthstone-dcg/audits/9},
  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: "dcg security audit report (audit version 9)"
version: "unspecified"
type: report
authors:
  - name: "Dicklesworthstone"
date-released: "2026-07-09"
url: "https://skillstore.io/skills/dicklesworthstone-dcg/audits/9"
identifiers:
  - type: other
    value: "skillstore:dicklesworthstone-dcg:audit:9"
    description: "Skillstore immutable audit report identifier"

Skillstore Score

Why this score Evidence Confidence: Medium

Architecture

Maintainability

Content

Community

Spec Compliance

What You Can Build

Protect active development work

Block accidental git resets, forced checkouts, and recursive deletion commands during agent-assisted coding sessions.

Add guardrails for agent teams

Standardize command safety checks across Claude Code sessions that run frequent Bash operations.

Audit risky shell command patterns

Review documented patterns for destructive database, container, cloud, and infrastructure commands before enabling packs.

Try These Prompts

Install the guard safely

Help me review the DCG installation steps before I run them. Explain what each command changes and which steps need extra caution.

Configure Claude Code hooks

Show me how to connect DCG to Claude Code with a PreToolUse Bash hook, and explain how I can test that it blocks git reset --hard.

Choose safety packs

Recommend a DCG pack configuration for a team that uses Git, Docker, Kubernetes, Terraform, and PostgreSQL. Explain the tradeoffs.

Evaluate command coverage

Review our risky shell command workflows and map them to DCG packs. Identify commands that DCG will not inspect and suggest compensating controls.

Best Practices

Review remote installation scripts before piping them to a shell.
Start with core protections, then enable extra packs that match tools your team uses.
Test hook behavior with harmless sample commands before relying on it in real projects.

Avoid

Do not treat the guard as protection against malicious users or compromised scripts.
Do not enable bypass variables in shared agent environments.
Do not run the sudo installation path unless system-wide installation is required and reviewed.

Frequently Asked Questions

Does this skill execute commands by itself?

The skill documentation does not execute commands, but it provides installation and test commands for users to run.

Why is the audit strict about the installer?

The quick install pipes remote content into a shell. That is convenient but risky unless the script is reviewed first.

Can DCG stop every destructive command?

No. It blocks known Bash patterns and allows unknown commands by default.

Does it protect non-Bash tools?

No. The documented hook targets Claude Code Bash tool calls.

Can users override a block?

Yes. The documentation describes manual execution and a bypass variable, so teams should restrict bypass use.

Is this useful with Codex or Claude?

The marketplace metadata lists Claude, Codex, and Claude Code, but the documented native hook integration is for Claude Code.

Developer Details

Author

Dicklesworthstone

License

MIT

Skillstore revision

Version notice

The author did not declare a version.

Repository

https://github.com/Dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations/tree/main/skills/dcg

Ref

1ffa7643651792ccb4bd3b15d924d2c97edff755

Maintenance freshness

7/18/2026

Usage

3 downloads · 340 views

File structure

📄 SKILL.md