Security - RTK (Rust Token Killer)

RTK is a CLI tool that executes shell commands and handles user input. This page covers security policies, the 3-layer review process, and best practices for contributors.

Reporting Vulnerabilities

If you discover a security vulnerability in RTK, please report it privately:

Email: security@rtk-ai.dev (or create a private security advisory on GitHub)
Response time: We aim to acknowledge reports within 48 hours
Disclosure: We follow responsible disclosure practices (90-day embargo)

Please do NOT:

Open public GitHub issues for security vulnerabilities
Disclose vulnerabilities on social media or forums before we’ve had a chance to address them

Security Threats

RTK faces several security concerns:

Shell injection: Command execution vulnerabilities
Supply chain attacks: Malicious dependencies
Backdoors: Logic bombs, exfiltration code
Data leaks: tracking.db exposure, telemetry abuse
Privilege escalation: Container/Docker operations

3-Layer Security Review Process

Layer 1: GitHub Action Security Checks

Every PR triggers automated security scanning via .github/workflows/security-check.yml:

Dependency audit

cargo audit - Detects known CVEs in dependencies

cargo audit

Fails the build if vulnerabilities are found.

Critical files alert

Flags modifications to high-risk files:

src/runner.rs - Shell execution engine
src/summary.rs - Command output aggregation
src/tracking.rs - SQLite database operations
src/pnpm_cmd.rs - Package name validation
src/container.rs - Docker/container operations
Cargo.toml - Dependency manifest
.github/workflows/*.yml - CI/CD pipelines

Dangerous pattern scan

Regex-based detection of suspicious code:

bash scripts/detect-dangerous-patterns.sh <PR_DIFF>

Scans for:

Shell execution (Command::new("sh"))
Environment manipulation (.env("LD_PRELOAD"))
Network operations (reqwest::, std::net::)
Unsafe code blocks
Panic-inducing patterns (.unwrap() in production)
Logic bombs (SystemTime::now() > ...)
Obfuscation (base64/hex strings)

Clippy security lints

cargo clippy enforces Rust best practices:

cargo clippy --all-targets

Catches unsafe patterns, performance issues, and code quality problems.

Results are posted in the PR’s GitHub Actions summary.

Layer 2: Claude Code Skill (Automated Review)

For maintainers with Claude Code access:

# Run the RTK PR security review skill
/rtk-pr-security <PR_NUMBER>

The skill performs comprehensive analysis:

Reviews all changed files for security issues
Checks for dangerous patterns with context
Validates dependency changes against crates.io
Identifies potential logic bombs and backdoors
Generates detailed security report

Benefits:

Faster review than manual inspection
Consistent security standards
Context-aware pattern detection
Reduces maintainer burden

Layer 3: Manual Security Review

For PRs modifying critical files or adding dependencies, maintainers perform manual review:

Review PR description

Verify intent matches implementation:

gh pr view <PR_NUMBER>

Inspect diff

Download and review the full diff:

gh pr diff <PR_NUMBER> > /tmp/pr.diff
less /tmp/pr.diff

Run dangerous pattern detection

bash scripts/detect-dangerous-patterns.sh /tmp/pr.diff

Audit new dependencies

For dependency additions in Cargo.toml:

Check crates.io page (downloads, maintainer, license)
Review GitHub repository (activity, issues, PRs)
Verify no typosquatting (similar names to popular crates)
Check for recent security advisories

Test locally

gh pr checkout <PR_NUMBER>
cargo build --release
./target/release/rtk <test-commands>

Complete review checklist

No critical files modified OR changes justified + reviewed by 2 maintainers
No dangerous patterns OR patterns explained + safe
No new dependencies OR deps audited on crates.io
PR description matches actual code changes
No logic bombs (time-based triggers, conditional backdoors)
Code quality acceptable (no unexplained complexity spikes)

Critical Files Requiring Enhanced Review

Tier 1: Shell Execution & System Interaction

src/runner.rs - Shell command execution engine (primary injection vector)
src/summary.rs - Command output aggregation (data exfiltration risk)
src/tracking.rs - SQLite database operations (privacy/telemetry concerns)

Tier 2: Input Validation

src/pnpm_cmd.rs - Package name validation (prevents injection via malicious names)
src/container.rs - Docker/container operations (privilege escalation risk)

Tier 3: Supply Chain & CI/CD

Cargo.toml - Dependency manifest (typosquatting, backdoored crates)
.github/workflows/*.yml - CI/CD pipelines (release tampering, secret exfiltration)

If your PR modifies ANY of these files, expect:

Detailed manual security review
Request for clarification on design choices
Potentially slower merge timeline
2-reviewer approval requirement

Dangerous Patterns

The security scanner checks for these patterns:

Pattern	Risk	Example
`Command::new("sh")`	Shell injection	Spawns shell with user input
`.env("LD_PRELOAD")`	Library hijacking	Preloads malicious shared libraries
`reqwest::`, `std::net::`	Data exfiltration	Unexpected network operations
`unsafe {`	Memory safety	Bypasses Rust’s guarantees
`.unwrap()` in `src/`	DoS via panic	Crashes on invalid input
`SystemTime::now() > ...`	Logic bombs	Delayed malicious behavior
Base64/hex strings	Obfuscation	Hides malicious URLs/commands

Example: Shell Injection

// ❌ DANGEROUS: Shell injection risk
let user_input = get_arg();
Command::new("sh")
    .arg("-c")
    .arg(format!("echo {}", user_input))  // User input in shell command!
    .output();

// ✅ SAFE: No shell, direct binary execution
let user_input = get_arg();
Command::new("echo")
    .arg(user_input)  // Passed as argument, not evaluated by shell
    .output();

Example: Logic Bomb

// ❌ DANGEROUS: Time-based trigger
use std::time::SystemTime;

if SystemTime::now() > SystemTime::UNIX_EPOCH + Duration::from_secs(1735689600) {
    // Malicious code executes after 2025-01-01
    std::fs::remove_dir_all("/");
}

// ✅ SAFE: No time-based conditions
// If you need time-based logic, make it explicit and documented

Security Best Practices

Error Handling

use anyhow::{Context, Result};

// ✅ GOOD: Graceful error handling
let path = std::env::args().nth(1)
    .context("Missing path argument")?;

// ❌ BAD: Panic on invalid input (DoS risk)
let path = std::env::args().nth(1).unwrap();

Input Validation

use regex::Regex;

// ✅ GOOD: Validate package names
fn validate_package_name(name: &str) -> Result<()> {
    let valid = Regex::new(r"^[a-z0-9-]+$").unwrap();
    if !valid.is_match(name) {
        anyhow::bail!("Invalid package name: {}", name);
    }
    Ok(())
}

// ❌ BAD: No validation
fn install_package(name: &str) -> Result<()> {
    Command::new("npm").arg("install").arg(name).output()?;
    Ok(())
}

Avoid Secrets in Code

// ❌ BAD: Hardcoded secrets
const API_KEY: &str = "sk_live_1234567890abcdef";

// ✅ GOOD: Environment variables or config files
let api_key = std::env::var("API_KEY")
    .context("API_KEY environment variable not set")?;

Exit Code Preservation

// ✅ GOOD: Preserve underlying tool's exit code
if !output.status.success() {
    let stderr = String::from_utf8_lossy(&output.stderr);
    eprintln!("{}", stderr);
    std::process::exit(output.status.code().unwrap_or(1));
}

// ❌ BAD: Always exit 0 (hides errors)
let output = Command::new("git").args(args).output()?;
println!("{}", String::from_utf8_lossy(&output.stdout));
Ok(())  // Returns success even if git failed!

Dependency Security

New dependencies added to Cargo.toml must meet these criteria:

Downloads: >10,000 on crates.io (or strong justification if lower)
Maintainer: Verified GitHub profile + track record of other crates
License: MIT or Apache-2.0 compatible
Activity: Recent commits (within 6 months)
No typosquatting: Manual verification against similar crate names

Red Flags

Brand new crate (<1 month old) with low downloads
Anonymous maintainer with no GitHub history
Crate name suspiciously similar to popular crate (e.g., serid vs serde)
License change in recent versions
Deprecated or unmaintained

Auditing Dependencies

# Check for known vulnerabilities
cargo audit

# View dependency tree
cargo tree

# Check for outdated dependencies
cargo outdated

# Update dependencies
cargo update

Disclosure Timeline

When vulnerabilities are reported:

Day 0: Acknowledgment sent to reporter
Day 7: Maintainers assess severity and impact
Day 14: Patch development begins
Day 30: Patch released + CVE filed (if applicable)
Day 90: Public disclosure (or earlier if patch is deployed)

Critical vulnerabilities (remote code execution, data exfiltration) may be fast-tracked.

Security Tooling

cargo audit - Automated CVE scanning (runs in CI)
cargo deny - License compliance + banned dependencies
cargo clippy - Lints for unsafe patterns
GitHub Dependabot - Automated dependency updates
GitHub Code Scanning - Static analysis via CodeQL (planned)

Contact

Security issues: security@rtk-ai.dev
General questions: https://github.com/rtk-ai/rtk/discussions
Maintainers: @FlorianBruniaux (active fork maintainer)

Last updated: 2026-02-02

​Reporting Vulnerabilities

​Security Threats

​3-Layer Security Review Process

​Layer 1: GitHub Action Security Checks

​Layer 2: Claude Code Skill (Automated Review)

​Layer 3: Manual Security Review

​Critical Files Requiring Enhanced Review

​Tier 1: Shell Execution & System Interaction

​Tier 2: Input Validation

​Tier 3: Supply Chain & CI/CD

​Dangerous Patterns

​Example: Shell Injection

​Example: Logic Bomb

​Security Best Practices

​Error Handling

​Input Validation

​Avoid Secrets in Code

​Exit Code Preservation

​Dependency Security

​Red Flags

​Auditing Dependencies

​Disclosure Timeline

​Security Tooling

​Contact

Reporting Vulnerabilities

Security Threats

3-Layer Security Review Process

Layer 1: GitHub Action Security Checks

Layer 2: Claude Code Skill (Automated Review)

Layer 3: Manual Security Review

Critical Files Requiring Enhanced Review

Tier 1: Shell Execution & System Interaction

Tier 2: Input Validation

Tier 3: Supply Chain & CI/CD

Dangerous Patterns

Example: Shell Injection

Example: Logic Bomb

Security Best Practices

Error Handling

Input Validation

Avoid Secrets in Code

Exit Code Preservation

Dependency Security

Red Flags

Auditing Dependencies

Disclosure Timeline

Security Tooling

Contact