Package methods as reusable Agent Skills
Requirements analysis, TestSpec design, API and cross-platform execution share reviewable contracts. A coding agent loads the appropriate workflow instead of improvising a new testing process for every task.
TestKit.
AI-Native Software Delivery & Verification System
Reusable Agent Skills and deterministic verification workflows that connect requirements, TestSpec, execution, defect regression and acceptance evidence.
- Area
- Verification
- Implementation focus
- Reusable Agent Skills, test specifications, execution contracts and deterministic validators.
- What you can inspect
- Explore the evidence workflow, including the conditions that keep an acceptance result open.
A verification chain you can inspect.
Select a stage to inspect its inputs, responsibilities, artifacts and validation rules.
From requirement to acceptanceInteractive guide
Requirement
testspec-new / testspec-update- Input
- Current requirements, product answers and acceptance rules.
- What the Agent Skill does
- Establish the product baseline, resolve scope and record questions that still need a human answer.
- Artifact
- Requirement document and change context.
- How it is verified
- Check requirement identity, version and authority. Unanswered questions remain explicit.
- Next dependency
- TestSpec consumes the confirmed requirement baseline, not an inferred implementation goal.
This is an explanation of the workflow, not a live test run. Defect / regression verification applies when a failure or repair needs investigation.
The problem
AI coding can produce changes quickly, but a completed task or green test does not explain which requirements were checked, on which build, or with what evidence. A reusable verification layer makes those relationships explicit.
How the pieces fit
Verification workflow
Requirement
TestSpec
Execution → Evidence
Regression (if needed) → Acceptance
Agent Skills · Deterministic validators
This diagram explains TestKit’s evidence workflow. Stages are inspectable concepts, not a claim that a live test run is happening in this browser. Defect / regression work is conditional on failures or repair verification.
Key engineering decisions
Freeze scope before execution
The source requirements or cases, build identity, environment and required checks form the execution baseline. Results retain every attempt so retries cannot erase a previous failure.
Keep product intent authoritative
Current requirements and confirmed product answers lead TestSpec. Code and historical cases are calibration evidence; they do not silently redefine the acceptance criteria.
Compute conclusions from evidence
Deterministic validators check coverage, source identity and current targets. A successful defect retest is distinct from a fresh acceptance run across all required checks.
Reliability & failure handling
Stale evidence stays historical
When requirements or the tested build change, earlier evidence is marked stale for the current target. The affected scope must be frozen and executed again.
Reproduce the actual failure
RED must demonstrate the defect signature. DNS, installation or account failures are blockers—not successful reproduction. GREEN and REGRESSION require comparable conditions.
Local checks have a local scope
The repository’s unified Python check runs packaging, contracts, synthetic evaluations and unit tests. Real devices, external APIs and full acceptance need their own authorized execution and evidence.
Evidence & scope
Source links are pinned to the reviewed revision so you can follow each implementation detail.
Acceptance contract
Frozen scope, all attempts, current targets and calculated conclusions.
skills/test-acceptance/SKILL.mdExecution evidence contract
Shared API, Web and mobile execution and evidence rules.
skills/_test-run-shared/references/execution-contract.mdDefect verification
RED / GREEN / REGRESSION and separate reacceptance.
skills/defect-verification/SKILL.mdSource provenance
Requirement authority and explicitly scoped code evidence.
skills/_testspec-shared/references/source-provenance.mdDeterministic validation entry
Packaging, contracts, synthetic eval definitions and unit checks.
scripts/test_all.pyWhat this case establishes
A delivery and verification capability layer for coding agents. Devices, business accounts and runtime tools are provided by the authorized host environment.