Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Validate Tests?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/gittower-validate-tests)
---
name: validate-tests
description: Codex-gate the test plan in an implementation plan before implementation starts
allowed-tools: Read, Grep, Glob, Edit, Write, Bash, Agent
---
# Validate Tests (Codex Gate)
Gate the test plan of an implementation plan before any implementation
starts. Combines a local check against the testing guidelines with an
external Codex review, following the shared convention in
[../_shared/CODEX_GATE.md](../_shared/CODEX_GATE.md).
This is a **required step** between `/create-plan` and `/implement`. Once
the plan passes this gate, the test plan is authoritative — `/implement`
treats it as immutable.
## Instructions
1. **Find the Plan**
- Detect current workflow folder from git branch
- Read `.ai/<folder>/plan.md`
- If no plan exists, suggest running `/create-plan` first
2. **Local Validation**
Load TESTING_GUIDELINES.md and GIT_TEST_SCENARIOS.md, then check the
Test Plan section:
- Each scenario follows the guidelines (setup patterns, assertions that
verify behavior rather than just "no error", proper Git scenario setup)
- Pay special attention to rules marked CRITICAL in the guidelines
- All code paths, error conditions, and edge cases have scenarios
- Scenarios are concrete enough to implement without guessing
Fix what you find directly in the plan.
3. **Codex Gate**
Run a Codex gate per [../_shared/CODEX_GATE.md](../_shared/CODEX_GATE.md):
- **Artifact**: the Test Plan section of `.ai/<folder>/plan.md`
- **Task for Codex**: find missing scenarios, wrong or underspecified
expected outcomes, and scenarios that don't match the source of truth.
TDD framing: these tests define the design — gaps found now are cheap,
gaps found during implementation are expensive
- **Source of truth**: the spec issue or `analysis.md`/`concept.md` the
plan was created from (pass its content or path)
- **Guidelines**: TESTING_GUIDELINES.md, GIT_TEST_SCENARIOS.md
4. **Evaluate and Apply**
Evaluate every Codex finding per the convention: apply only with high
confidence, reject over-engineering and hypothetical scenarios, when in
doubt leave it out. Update the plan's Test Plan section with applied
findings.
Log all findings — applied, partial, and rejected with reasons — to
`.ai/<folder>/codex-test-plan.md`.
5. **Report Findings**
Summarize:
- Local validation: issues found and fixed
- Codex gate: findings applied / rejected (with the log path)
- Open questions that need the user (genuine design decisions only)
- Confirmation that the plan is gated and ready for `/implement`