Updates, generates, and validates tests using git-workspace context and TDD/BDD methodology. Use when code changes require new or updated test coverage.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Updates, generates, and validates tests using git-workspace context and TDD/BDD methodology. Use when code changes require new or updated test coverage.
detailed test management system that applies TDD/BDD principles to maintain, generate, and enhance tests across codebases. This skill practices what it preaches - it uses TDD principles for its own development and serves as a living example of best practices.
Core Philosophy
RED-GREEN-REFACTOR: Strict adherence to TDD cycle
Behavior-First: BDD patterns that describe what code should do
Invariant-Encoding: Tests guard design decisions, not just behavior
Meta Dogfooding: The skill's own tests demonstrate the principles it teaches
Quality Gates: detailed validation before considering tests complete
What It Is
A modular test management system that:
Discovers what needs testing or updating
Generates tests following TDD principles
Enhances existing tests with BDD patterns
Validate test quality through multiple lenses
Quick Start
Quick Checklist for First Time Use
validate pytest is installed (pip install pytest)
Have your source code in src/ or similar directory
Create a tests/ directory if it doesn't exist
Run Skill(sanctum:git-workspace-review) first to understand changes
Start with Skill(test-updates) --target <specific-module> for focused updates
detailed Test Update
# Run full test update workflow
Skill(test-updates)
Verification: Run pytest -v to verify tests pass.
Targeted Test Updates
# Update tests for specific paths
Skill(test-updates) --target src/sanctum/agents
Skill(test-updates) --target tests/test_commit_messages.py
Verification: Run pytest -v to verify tests pass.
TDD for New Features
# Apply TDD to new code
Skill(test-updates) --tdd-only --target new_feature.py
Verification: Run pytest -v to verify tests pass.
Using the Scripts Directly
Human-Readable Output:
# Analyze test coverage gaps
python plugins/sanctum/scripts/test_analyzer.py --scan src/
# Generate test scaffolding
python plugins/sanctum/scripts/test_generator.py \
--source src/my_module.py --style pytest_bdd
# Check test quality
python plugins/sanctum/scripts/quality_checker.py \
--validate tests/test_my_module.py
See modules/test-discovery.md for detection patterns.
Phase 2: Strategy
Choose appropriate BDD style (see modules/bdd-patterns.md)
Plan test structure
Define quality criteria
Identify design invariants to encode as tests
Phase 2.5: Invariant-Encoding Tests
Before writing behavioral tests, identify the design
invariants that the code relies on and write tests
that would break if those invariants were violated.
What to encode:
Module boundary constraints (A never imports from B)
Data flow direction (events flow publisher-to-subscriber,
never the reverse)
API contract shapes (public interfaces don't change
without versioning)
Data structure choices (if a map was chosen over a list,
test the properties that justify that choice)
deftest_plugins_never_import_from_other_plugins():
"""Encode the invariant: plugins are independent modules.
If this test breaks, someone is coupling plugins
directly. Present the 3 options to a human:
1. Preserve: revert the import, keep plugins independent
2. Layer: add a shared interface in leyline instead
3. Revise: merge the plugins (requires ADR)
"""for plugin_dir in plugin_dirs:
imports = extract_imports(plugin_dir)
for imp in imports:
assertnot imp.startswith("plugins."), (
f"{plugin_dir} imports {imp} — "f"violates plugin independence invariant"
)
Why this matters: Tests that encode invariants are
load-bearing. When an agent later encounters a feature
that clashes with the invariant, the test failure forces
a conscious decision rather than a silent drift. Without
these tests, bad invariant decisions compound until the
codebase is unsalvageable.
When updating existing tests:
If an invariant-encoding test needs to change, do NOT
silently update the assertion. Flag it for human review
with the three options: preserve the invariant, layer
on top, or revise the invariant. This is a judgment
call that requires human wisdom: models default to
the "average" of training data and get these wrong far
too often.
Phase 3: Implementation
Write failing tests (RED) - see modules/tdd-workflow.md
Implement minimal passing code (GREEN)
Refactor for clarity (REFACTOR)
See modules/test-generation.md for generation templates.
Phase 4: Validation
Static analysis and linting
Dynamic test execution
Coverage and quality metrics
See modules/quality-validation.md for validation criteria.
Quality Assurance
The skill applies multiple quality checks:
Static: Linting, type checking, pattern validation
Invariant: Verify design-decision tests are not weakened
Review: Structured checklists for peer validation
Examples
BDD-Style Test Generation
See modules/bdd-patterns.md for additional patterns.
classTestGitWorkflow:
"""BDD-style tests for Git workflow operations."""deftest_commit_workflow_with_staged_changes(self):
"""
GIVEN a Git repository with staged changes
WHEN the user runs the commit workflow
THEN it should create a commit with proper message format
AND all tests should pass
"""# Test implementation following TDD principlespass
Verification: Run pytest -v to verify tests pass.
Test Enhancement
Add edge cases and error scenarios
Include performance benchmarks
Add mutation testing for robustness
See modules/test-enhancement.md for enhancement strategies.
Q: Tests are failing after generation
A: This is expected! The skill follows TDD principles - generated tests are designed to fail first. Follow the RED-GREEN-REFACTOR cycle:
Run the test and confirm it fails for the right reason
Implement minimal code to make it pass
Refactor for clarity
Q: Quality score is low despite having tests
A: Check for these common issues:
Missing BDD patterns (Given/When/Then)
Vague assertions like assert result is not None
Tests without documentation
Long, complex tests (>50 lines)
Q: Generated tests don't match my code structure
A: The scripts analyze AST patterns and may need guidance:
Use --style flag to match your preferred BDD style
Check that source files have proper function/class definitions
Review the generated scaffolding and customize as needed
Q: Mutation testing takes too long
A: Mutation testing is resource-intensive:
Use --quick-mutation flag for subset testing
Focus on critical modules first
Run overnight for detailed analysis
Q: Can't find tests for my file
A: The analyzer uses naming conventions:
Source: my_module.py → Test: test_my_module.py
Check that test files follow pytest naming patterns
validate test directory structure is standard
Performance Tips
Large codebases: Use --target to focus on specific directories
CI integration: Run validation in parallel with other checks
Memory usage: Process files in batches for very large projects
Getting Help
Check script outputs for detailed error messages
Use --verbose flag for more information
Review the validation report for specific recommendations
Start with small modules to understand patterns before scaling
Exit Criteria
pytest -v passes with zero failures after all test updates
are applied to the target files
Test coverage for files in scope exceeds 85% as reported by
pytest --cov
All new tests include a GIVEN/WHEN/THEN docstring matching
the BDD pattern from modules/bdd-patterns.md
quality_checker.py --validate <test_file> --output-json
returns quality_score ≥ 80 for each updated test file
If an invariant-encoding test changes, it is flagged for human
review with the three options (preserve/layer/revise) before
any assertion is modified
