| description | Security practices for the B2C Developer Tooling project including supply chain protections and dependency management. |
|---|
This page covers security practices used in the B2C Developer Tooling project, with a focus on supply chain security.
The JavaScript/Node.js ecosystem is particularly vulnerable to supply chain attacks due to the large number of transitive dependencies in typical projects. This project uses several pnpm features to mitigate these risks.
New package versions are quarantined for 48 hours before they can be installed:
# pnpm-workspace.yaml
minimumReleaseAge: 2880 # minutes (48 hours)This provides a buffer period during which:
- Malicious packages can be detected and removed from npm
- Security researchers can identify and report compromised packages
- The community can flag suspicious updates
If a package update is urgent, it can be added to the exclusion list:
minimumReleaseAgeExclude:
- some-urgent-packageDependency downgrades are prevented to protect against downgrade attacks:
# pnpm-workspace.yaml
trustPolicy: no-downgradeThis ensures that once a package version is installed, it cannot be replaced with an older (potentially vulnerable) version without explicit action.
Only explicitly allowed packages can run build scripts (install/postinstall hooks):
# pnpm-workspace.yaml
onlyBuiltDependencies:
- unrs-resolver
- yarnBuild scripts are a common attack vector because they execute arbitrary code during installation. By default, pnpm blocks all build scripts except for packages in this allowlist.
When adding a new dependency that requires build scripts:
- Verify the package is legitimate and actively maintained
- Review what the build script does
- Add it to
onlyBuiltDependenciesif necessary
This project uses NPM trusted publishers for package publication. Instead of storing long-lived npm tokens, packages are published via GitHub Actions using short-lived OIDC tokens that cannot be extracted or reused.
The CLI includes a Safety Mode feature that prevents accidental or unwanted destructive operations. This is particularly important when:
- Using the CLI in automated environments (CI/CD pipelines)
- Providing the CLI as a tool to AI agents/LLMs
- Working in production environments
- Training new team members
- Running commands from untrusted scripts
Safety Mode uses a hybrid protection approach:
-
HTTP Middleware Layer (Primary Protection)
- Intercepts ALL HTTP requests before they're sent
- Cannot be bypassed by command-line flags
- Works automatically for all commands
- LLM-proof: controlled via environment variable
-
Command-Level Checks (Better UX)
- Provides early, user-friendly error messages
- Catches operations before HTTP requests
Configure via the SFCC_SAFETY_LEVEL environment variable:
| Level | Description | Blocks |
|---|---|---|
NONE |
No restrictions (default) | Nothing |
NO_DELETE |
Prevent deletions | DELETE operations |
NO_UPDATE |
Prevent deletions and destructive updates | DELETE + reset/stop/restart |
READ_ONLY |
Read-only mode | All writes (POST/PUT/PATCH/DELETE) |
# No restrictions (default)
unset SFCC_SAFETY_LEVEL
# OR
export SFCC_SAFETY_LEVEL=NONE
b2c sandbox delete test-id # ✅ Allowed# Prevent accidental deletions in automated environments
export SFCC_SAFETY_LEVEL=NO_DELETE
b2c sandbox create --realm test # ✅ Allowed
b2c sandbox delete test-id # ❌ Blocked# Prevent AI agents from performing destructive operations
export SFCC_SAFETY_LEVEL=NO_UPDATE
b2c sandbox list # ✅ Allowed
b2c sandbox create --realm test # ✅ Allowed
b2c sandbox delete test-id # ❌ Blocked
b2c sandbox reset test-id # ❌ Blocked# Absolute read-only mode
export SFCC_SAFETY_LEVEL=READ_ONLY
b2c sandbox list # ✅ Allowed
b2c sandbox get test-id # ✅ Allowed
b2c sandbox create test # ❌ Blocked| Safety Level | DELETE | POST (create) | POST (reset) | PUT/PATCH | GET |
|---|---|---|---|---|---|
| NONE | ✅ | ✅ | ✅ | ✅ | ✅ |
| NO_DELETE | ❌ | ✅ | ✅ | ✅ | ✅ |
| NO_UPDATE | ❌ | ✅ | ❌ | ✅ | ✅ |
| READ_ONLY | ❌ | ❌ | ❌ | ❌ | ✅ |
Safety Mode automatically protects ALL destructive commands across all topics:
- Sandbox:
delete,reset,alias delete - Account Manager:
users delete,users reset,clients delete - Code:
delete - MRT:
project delete,env delete,env var delete,env redirect delete,project notification delete - SLAS:
client delete - eCDN: All delete operations (certificates, zones, rules, etc.)
Environment variables are more secure than command-line flags because:
- LLMs Don't Control Them: When an LLM uses the CLI as a tool, it controls commands and flags but NOT the environment
- Session-Level: Set once for the entire session/container
- Audit Trail: Can be logged and monitored in CI/CD
- Cannot Be Bypassed: Even
--forceflags don't override safety mode
When an operation is blocked, you'll see clear error messages:
export SFCC_SAFETY_LEVEL=NO_DELETE
b2c sandbox delete test-id
# Error: Cannot delete sandbox: blocked by safety level NO_DELETE.
#
# Delete operations are blocked in NO_DELETE mode
#
# To allow this operation, unset or change the SFCC_SAFETY_LEVEL environment variable.# GitHub Actions example
- name: Deploy to Production
env:
SFCC_SAFETY_LEVEL: NO_DELETE # Prevent accidental deletions
run: |
b2c code deploy
b2c sandbox start production# Provide CLI to LLMs with safety enabled
export SFCC_SAFETY_LEVEL=NO_UPDATE
# LLMs can now read and create, but cannot delete or reset# Set in shell profile for production access
echo 'export SFCC_SAFETY_LEVEL=NO_UPDATE' >> ~/.bashrcVerify safety mode is working:
export SFCC_SAFETY_LEVEL=NO_DELETE
b2c sandbox delete fake-id
# Expected: "blocked by safety level NO_DELETE"
# NOT expected: Authentication error or API callFor comprehensive testing, see GitHub Issue #67.
- Review dependency updates carefully, especially for packages with build scripts
- Be cautious when adding new dependencies
- Prefer packages with minimal transitive dependencies
- Check package health on npm before adding (download counts, maintenance activity, known vulnerabilities)
- Keep the CLI updated to receive security patches
- Review the
pnpm-workspace.yamlsettings if you fork or modify this project - Consider using similar protections in your own projects
- Use Safety Mode when running CLI in automated environments or providing it as a tool to AI agents