create-test-scenarios

name: create-test-scenarios description: Create Automatus test scenarios to test the given rule. argument-hint: ""

Create Test Scenarios

Create test scenarios for execution by the Automatus test framework. Automatus and test scenarios are documented in tests/README.md.

Scope: This skill covers Linux rules only. It does not cover OpenShift tests (e.g., tests in ocp4 directories).

Phase 1: Understand the Rule

Goal: Produce a summary of key information about the rule.

1. Locate the rule directory containing the rule.yml file. The directory name matches the rule ID.
2. Read rule.yml.
   • Focus on the description and rationale fields to identify what to test.
   • Determine whether the rule is templated (templated rules contain a template key in rule.yml)
3. Analyze rendered content (OVAL checks and remediations).
   • Call mcp__content-agent__get_rendered_rule(product, rule_id) to see the actual OVAL check and remediation code after Jinja2 expansion. This is the most important tool for understanding what to test -- the raw rule.yml does not show final file paths, config keys, or values after template expansion.
   • Identify exactly which files, directories, config options, and packages the OVAL check examines.
   • Identify exactly how the Bash and Ansible remediation scripts modify those items.
   • Test scenarios must exercise the exact paths and values from the rendered content.
4. Determine target product
   • Use mcp__content-agent__get_rule_product_availability(rule_id) to see which products include the rule.
5. Find existing and similar tests.
   • Read any existing test scenarios in the rule's tests/ subdirectory.
   • Read test scenarios for similar rules (same template or same subsystem) to match conventions and avoid reinventing patterns.
   • For templated rules, read existing templated test scenarios in shared/templates/${template_name}/tests.
6. Review shared helpers and macros.
   • Browse tests/shared/ to discover reusable Bash helper scripts (see "Shared test helpers" below).
   • Read shared/macros/20-test-scenarios.jinja to discover available Jinja macros for test setup (see "Jinja macros for test scenarios" below).
   • Prefer existing helpers and macros over custom setup code.

MCP tools to use:

• mcp__content-agent__get_rule_details(rule_id) -- quickly identifies template name and metadata
• mcp__content-agent__get_rendered_rule(product, rule_id) -- shows the actual OVAL check and remediation code after Jinja2 expansion
• mcp__content-agent__search_rendered_content(query, product, limit) -- searches rendered build artifacts (useful for finding actual values after template expansion)
• mcp__content-agent__get_template_schema(template_name) -- for templated rules, shows parameter schema

Phase 2: Design Test Scenarios

Goal: Produce a design document describing the test scenarios to create.

Use the information gathered in Phase 1 to propose test scenarios.

Pass and fail scenarios

Propose a set of "pass" and "fail" test scenarios:

• Pass scenarios configure the system to a compliant state. They do not test remediations, because a passing system does not need remediation.
• Fail scenarios configure the system to a non-compliant state. Automatus runs the fail script, scans (expects fail), remediates automatically, then rescans (expects pass). Therefore, a .fail.sh scenario must put the system in a state that is both detectable as non-compliant by the OVAL check and fixable by the remediation.
  • Before designing a fail scenario, verify that the remediation can actually fix the state you are creating. For example, if the remediation only appends a line but does not modify existing lines, a fail scenario with a wrong value (rather than a missing value) may not be fixable.

Do not create "notapplicable" test scenarios. They are valid but serve specialized purposes outside this skill's scope.

Relationship between pass and fail scenarios

Pass and fail scenarios for the same rule typically share roughly 80% of their setup code. Fail scenarios differ from pass scenarios by:

• Omitting a critical setup step (e.g., not enabling a service)
• Setting a wrong or opposite value
• Removing or commenting out correct configuration
• Using a minimal setup to prove absence of required config

Design pass/fail pairs together to ensure consistency.

How Automatus executes test scenarios

• pass.sh: execute script → scan → expect PASS (done)
• fail.sh: execute script → scan → expect FAIL → remediate → rescan → expect PASS
• notapplicable.sh: execute script → scan → expect NOTAPPLICABLE (done)

How many scenarios to write

Every rule must have at least 1 pass and at least 1 fail test scenario. A typical rule has 5 to 10 test scenarios, but the exact number depends on the rule's complexity.

Suggested distribution:

• Simple rules (single file, single option): 2-3 pass, 2-3 fail
• Medium rules (multiple locations, .d directories): 2-3 pass, 4-5 fail
• Complex rules (multiple subsystems, product-specific logic): 3-4 pass, 5-7 fail

Typical situations to test

• Configuration option is absent
• Configuration option is commented out
• Configuration option has an incorrect value
• Multiple configuration options with different values
• Options in root configuration files
• Options in .d directories (test with and without files in the .d directory)
• File exists but is empty
• Correct value in wrong location (e.g., in a .d file when only the main file is checked, or vice versa)
• Runtime value differs from config file value (e.g., sysctl)

Focus on edge cases when designing test scenarios.

Phase 3: Write Tests

Goal: Produce executable Bash test scenario files based on the Phase 2 design.

Every scenario file must start with #!/bin/bash.

Jinja2 in test scenarios

Use Jinja2 macros and expressions in scenario files. This project uses non-standard delimiters:

• {{{ }}} for expressions
• {{% %}} for control flow

Use product properties such as {{{ grub2_boot_path }}} and {{{ audit_watches_style }}} to parametrize test scenarios across products.

Use product-specific conditionals when different products require different setup:

{{% if product in ["rhel9", "rhel10"] %}}
grubby --update-kernel=ALL --args="audit=1"
{{% elif "ubuntu" in product %}}
sed -i 's/GRUB_CMDLINE_LINUX="/&audit=1 /' /etc/default/grub
update-grub
{{% else %}}
grub2-editenv - set "$(grub2-editenv - list | grep kernelopts) audit=1"
{{% endif %}}

Additional Jinja macros for test scenarios are available in shared/macros/20-test-scenarios.jinja (see section below).

Header

Each test scenario file begins with a comment header (lines starting with #). Include only the headers that apply:

• packages -- packages that must be installed before the test runs
• platform -- restricts the scenario to specific platforms
• check -- use only when the rule has both OVAL and SCE checks but the scenario works with only one check engine
• remediation -- set to none when the rule has no remediation or when remediation would break the test environment
• variables -- sets XCCDF Value parameters; format is variable_name=actual_value (use the actual value, not a selector name)
• profiles -- use only for profile-specific regression tests where the rule is not parametrized by an XCCDF Value

File placement

Non-templated rules: Place test scenario files in the tests/ subdirectory of the rule directory. Create the directory if it does not exist.

Templated rules: Determine whether the scenario is specific to a single rule or can be parametrized and reused by other rules using the same template. Reusable scenarios are far more common.

• Rule-specific scenarios: Place in the rule's tests/ subdirectory.
• Reusable scenarios: Place in shared/templates/${template_name}/tests/.

In reusable test scenarios, use template parameters instead of specific values. Call mcp__content-agent__get_template_schema() to get the list of available parameters. Template parameters are substituted by Jinja. Write template parameter names in CAPITAL letters.

Controlling templated scenarios with test_config.yml

A rule can include tests/test_config.yml to control which templated scenarios run. This is important when some templated tests do not apply to a specific rule. The file supports Jinja2 and is product-aware.

Use deny_templated_scenarios to block specific templated scenarios:

deny_templated_scenarios:
  - wrong_runtime.fail.sh
  - missing_config.fail.sh

Use allow_templated_scenarios to allow only specific templated scenarios (all others are blocked). Use - none to disable all templated scenarios for the rule:

allow_templated_scenarios:
  - none

Shared test helpers

Source helper scripts from tests/shared/ using the $SHARED variable (e.g., . $SHARED/partition.sh).

Jinja macros for test scenarios

Macros from shared/macros/20-test-scenarios.jinja:

Specialized composition macros for specific rsyslog rules also exist (e.g., setup_rsyslog_encrypt_offload_actionsendstreamdriverauthmode()).

Tips for writing test code

• Create directories before writing files to them.
• Clean up unwanted values from configuration files before inserting correct values.
• Clean up all files from .d directories.
• When setting sysctl values, set both the config file (/etc/sysctl.conf or /etc/sysctl.d/) and the runtime value (sysctl -w key=value).
• When cleaning sysctl config, remove from all locations: /etc/sysctl.conf, /etc/sysctl.d/*, /usr/lib/sysctl.d/*, /run/sysctl.d/*.
• Use sed -i '/pattern/d' to remove existing config lines before adding new ones to prevent duplicate entries.
• Put each test scenario in a separate file with a short but descriptive name.
• File naming convention:
  • Pass scenarios: file name ends with .pass.sh
  • Fail scenarios: file name ends with .fail.sh

Examples

Example 1: Basic fail scenario (wrong_value.fail.sh)

#!/bin/bash
# packages = openssh-server
# platform = multi_platform_rhel

# Set the sshd option to a wrong value so the rule fails
sed -i '/KerberosAuthentication/d' /etc/ssh/sshd_config
echo "KerberosAuthentication yes" >> /etc/ssh/sshd_config

Example 2: Basic pass scenario (correct.pass.sh)

#!/bin/bash
# platform = Ubuntu 24.04

getent group "adm" &>/dev/null || groupadd adm
mkdir -p /var/log/apt
touch /var/log/apt/testfile
chgrp adm /var/log/apt/testfile

Example 3: Scenario with advanced features (augenrules_correct_extra_permission.pass.sh)

This example demonstrates:

• packages header to ensure the audit RPM package is installed
• platform header limiting applicability to RHEL products (rhel8, rhel9, rhel10)
• variables header setting an actual value (not a selector) for the XCCDF Value var_accounts_passwords_pam_faillock_dir
• Jinja macro setup_auditctl_environment() generating Bash code
• Product property audit_watches_style substituted into the code
• Sourcing a shared Bash script via the $SHARED variable

#!/bin/bash
# packages = audit
# platform = multi_platform_rhel
# variables = var_accounts_passwords_pam_faillock_dir=/var/log/faillock

{{{ setup_auditctl_environment() }}}

path="/var/log/faillock"
style="{{{ audit_watches_style }}}"
filter_type="path"
. $SHARED/audit_rules_watch/auditctl_correct_without_key.pass.sh

Example 4: Templated scenario for service_enabled template (service_disabled.fail.sh)

Template parameters are substituted by Jinja and must be written in CAPITAL letters.

#!/bin/bash
{{% if SERVICENAME in ["ssh", "sshd"] %}}
# platform = Not Applicable
{{% endif %}}
# packages = {{{ PACKAGENAME }}}

SYSTEMCTL_EXEC='/usr/bin/systemctl'
"$SYSTEMCTL_EXEC" stop '{{{ DAEMONNAME }}}.service'
"$SYSTEMCTL_EXEC" disable '{{{ DAEMONNAME }}}.service'

Example 5: Sysctl scenario (correct_value.pass.sh)

Demonstrates proper cleanup of all sysctl config locations and setting both config and runtime values:

#!/bin/bash
# variables = sysctl_{{{ SYSCTLID }}}_value={{{ SYSCTL_CORRECT_VALUE }}}

rm -rf /usr/lib/sysctl.d/* /run/sysctl.d/* /etc/sysctl.d/*
sed -i "/{{{ SYSCTLVAR }}}/d" /etc/sysctl.conf

echo "{{{ SYSCTLVAR }}} = {{{ SYSCTL_CORRECT_VALUE }}}" >> /etc/sysctl.conf
sysctl -w {{{ SYSCTLVAR }}}="{{{ SYSCTL_CORRECT_VALUE }}}"

Example 6: Mount option test using shared helpers (fstab.fail.sh)

#!/bin/bash
# platform = multi_platform_all

. $SHARED/partition.sh

clean_up_partition {{{ MOUNTPOINT }}}
create_partition
make_fstab_given_partition_line {{{ MOUNTPOINT }}} ext2 nodev
mount_partition {{{ MOUNTPOINT }}} || true

Example 7: Dconf test using shared helpers (correct_value.pass.sh)

#!/bin/bash
# packages = dconf,gdm

. $SHARED/dconf_test_functions.sh

clean_dconf_settings
add_dconf_setting "{{{ SECTION }}}" "{{{ PARAMETER }}}" "{{{ VALUE }}}" "{{{ DCONF_DATABASE_DIRECTORY }}}" "00-security-settings"
add_dconf_lock "{{{ SECTION }}}" "{{{ PARAMETER }}}" "{{{ DCONF_DATABASE_DIRECTORY }}}" "00-security-settings-lock"
dconf update

Phase 4: Verify by Running Automatus Tests

Run the created test scenarios using Automatus. Use the test-rule skill. Verify that all test scenarios pass. Run tests with both oscap and ansible remediations. Don't run the test with the bash remediations.

How to interpret Automatus results

For fail test scenarios, don't check only the initial scan results, watch for failures in the scan after remediation.

Example 1: Pass test scenario - PASS

The rule returned the expected result pass result during the initial scan.

INFO - Script banner_etc_issue_disa_dod_short.pass.sh using profile (all) OK

Example 2: Pass test scenario - FAIL

The rule didn't return the expected result pass result during the initial scan, instead it returned the fail result.

ERROR - Script banner_etc_issue_disa_dod_short.pass.sh using profile (all) found issue:
ERROR - Rule evaluation resulted in fail, instead of expected pass during initial stage 
ERROR - The initial scan failed for rule 'xccdf_org.ssgproject.content_rule_banner_etc_issue'.

Example 3: Fail test scenario - PASS

OpenSCAP evaluated the rule as fail during the first run, which was expected by the test scenario. Then, the remediation was executed, and after the remediation, the rule was evaluated as pass so oscap returned the fixed result in the second run.

INFO - Script banner_etc_issue_disa_dod_short.fail.sh using profile (all) OK

Example 4: Remediation doesn't work correctly - FAIL

A "fail" test scenario returned the expected fail results in the initial scan. Then, it executed the remediation, but after the remediation the rule was evaluated as fail again, which means that the scanner reported the error result for the rule. This signalizes that the remediation for this rule doesn't work correctly.

INFO - Script banner_etc_issue_disa_dod_short.fail.sh using profile (all) OK
ERROR - Rule evaluation resulted in error, instead of expected fixed during remediation stage 
ERROR - The remediation failed for rule 'xccdf_org.ssgproject.content_rule_banner_etc_issue'.

Common Pitfalls

• Forgetting to clean up .d directories before setting up the test state
• Writing fail scenarios that the remediation cannot fix (always verify that the remediation logic can handle the state you create)
• Writing tests that check paths or values not matching what the OVAL actually checks (always base tests on rendered content, not assumptions)
• Using the profiles header when the variables header should be used instead
• Using a selector name instead of the actual value in the variables header (e.g., # variables = var_name=selector instead of # variables = var_name=10)
• Using default Jinja syntax ({{ }}, {% %}) instead of the project's custom syntax ({{{ }}}, {{% %}})
• Missing shebang (#!/bin/bash)
• Using lowercase letters for template parameters in templated test scenarios
• Not setting the runtime value alongside the config file value (e.g., forgetting sysctl -w after editing sysctl.conf)
• Recreating shared helpers or audit watch scenarios instead of sourcing them from $SHARED