- Home
- Skills
- Testing & Debugging
- Test First Slices
Works with the AI Agents you already use
Test First Slices
Drives behavior with tests, one bounded slice at a time: prove red, make green, refactor under protection. No implementation until the red signal is trustworthy, and no simulated green when tests cannot run.
$18
Test First Slices
Example session with this skill installed
TDD the discount rule: 10% off orders over $100.
- Read your context and instructions
- Compiled the test first slices
Illustrative example, synthetic data. File names, findings and outputs below are made up to show the shape of a real run. In actual use the skill reports only evidence supplied or verified in your environment and marks anything unavailable as UNVERIFIED.
SLICE: order total > $100 gets 10% discount.
RED EVIDENCE: test "applies 10% to $120 -> $108" fails (no discount logic); fails for the intended reason, not setup noise.
IMPLEMENTED BEHAVIOR: applyDiscount(total) returns total*0.9 when total>100 else total.
GREEN EVIDENCE: that test passes; boundary test "$100 -> no discount" added and passes.
REFACTOR: extracted threshold/rate as named constants while green.
NEXT SLICE / STOP: next slice = stacking with coupon codes (belongs to approved objective); or stop if out of scope.
Connects securely to your tools. The creator never sees your data.
About this skill
Test First Slices uses executable expectations to constrain behavior one stable slice at a time. It selects one observable behavior, writes or adjusts the expectation, confirms it fails for the intended missing behavior and not for setup noise, implements the smallest coherent behavior, and refactors only while the relevant suite stays green. When tests cannot run, it marks the slice blocked instead of simulating a green test.
The problem it addresses
The skill's completion criterion is a suite that protects behavior rather than mirroring implementation details. A slice counts as done only if a meaningful failing expectation came first, the slice ends green at the appropriate test layer, and the next slice stays inside the approved objective.
How it works
It runs five phases in order.
- Select one behavior: translate the next acceptance criterion into one observable outcome and choose the lowest test layer that proves it without coupling to internals.
- Prove red: write or adjust the expectation and run it, then confirm it fails for the intended missing behavior and not for syntax or setup noise. Gate: no implementation until the red signal is trustworthy.
- Make green: implement the smallest coherent production behavior that satisfies the expectation while respecting architecture, without hard-coding only to the test input.
- Refactor under protection: remove duplication or improve names and seams only while the relevant suite remains green, without widening the change to unrelated cleanup.
- Repeat or stop: take the next slice only if it belongs to the approved objective, and escalate if the new test exposes a missing product or architecture decision.
Choosing the seam
Pure or domain behavior gets a focused unit test. A module contract gets a component or integration test. An HTTP or API contract gets a request and response test. User-visible interaction uses browser or end-to-end tests only when lower seams cannot prove it.
What you get back
When explicitly reported, a six-field report: SLICE, RED EVIDENCE, IMPLEMENTED BEHAVIOR, GREEN EVIDENCE, REFACTOR and NEXT SLICE / STOP. The package includes a filled-in worked example for a discount rule: a failing test, the smallest implementation, a green test with a boundary case, a refactor that extracts named constants, and a next slice.
Boundaries
It is not for throwaway prototypes, pure visual polish, generated configuration, or environments with no meaningful automated seam. If the normal test runner or target environment is unavailable, it uses a smaller executable seam only when that proves the same behavior; otherwise the slice is marked blocked. It touches only the approved surface and never commits, pushes, deploys or installs anything on its own. Project governance documents, ADRs and approved specs outrank the skill, and it stops at its handoff boundary.
How it differs from nearby work
The ordering is the mechanism. Nothing is implemented before a trustworthy red signal, so it is not a pass that writes tests after the code. Each pass is one bounded slice inside the approved objective, so it is not a broad implementation pass, and a test that exposes a missing product or architecture decision is escalated instead of resolved silently.
Evidence discipline
Facts, inferences and unknowns are kept separate, and a passing narrow check counts as proof only of the surface it exercised. The skill never assumes a shell, Git, browser, network or MCP tool exists, verifies capability first, and never claims a tool ran when it did not.
Validation scope
The package is five text files: the process, a README, a worked example and two reference documents. It names no specific language, framework or test runner, and ships no executable scripts, test suite or benchmark results. This listing makes no claim about measured outcomes or host-specific testing.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.
Test First Slices is for deterministic behavior with a stable observable seam. It is not meant for throwaway prototypes, pure visual polish, generated configuration or environments with no meaningful automated seam. It needs a way to run tests: without one the slice is marked blocked rather than passed. It guides one bounded slice at a time and does not guarantee model output or replace project-specific rules. The package ships no executable checks, benchmarks or language-specific test results, and a passing test proves only the behavior it exercises.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 17 days ago
Permissions
Required: read access to the repository files, tests and test output under work; write access to add or adjust the expectation and implement the behavior under test; and a shell or test runner to run the expectations, because the red gate and green evidence depend on running them. Optional: browser or end-to-end tests, used only when lower seams cannot prove the behavior. Where no runner is available, the skill uses a smaller executable seam or marks the slice blocked. The package names no network access, environment variables or external services. The skill never commits, pushes, deploys or installs anything on its own.