A skill for coding agents that verifies software by building and running it, operating its real interfaces, exploring varied inputs, and preserving reproducible failures.
Build it. Run it. Operate the changed feature. Explore how it fails. Preserve what you discover.
Install with the GitHub CLI:
gh skill install Corvidae-Coding-Projects/real-testing real-testingOr copy skills/real-testing/ into your agent's skills directory. Keep
SKILL.md, agents/, and assets/ together. The agent needs access to the
program's build tools, runtime dependencies, and interface.
For an agent that supports named skills, ask:
Use $real-testing to verify this change. Build and run the program, operate
the changed feature through its real interface, explore boundary and invalid
inputs, and preserve any confirmed failures as replayable regressions.
You can also give an agent SKILL.md directly and ask it to follow the workflow.
- Define the contract. Specify observable behavior and expected results before executing checks.
- Build and launch. Run the current version with its actual dependencies and isolated test data.
- Operate the feature. Exercise the interface its users or callers use.
- Explore systematically. Generate varied inputs or action sequences and check them against independent expectations or justified invariants.
- Fix and repeat. Reproduce confirmed defects, fix them, rebuild, and rerun the affected checks.
- Preserve failures. Keep concrete inputs, expected and observed behavior, prerequisites, and a command or procedure that detects recurrence.
- Report evidence. State what ran, what passed or failed, what was blocked, and the limits of the exploration.
| Software | How it is exercised |
|---|---|
| Command-line programs | Invoke the executable; check output, exit status, and artifacts. |
| APIs and services | Send real requests; inspect responses and downstream effects. |
| Web and desktop applications | Operate the changed controls; inspect displayed and saved state. |
| Libraries | Build and run a consumer of the public API. |
| Workers and pipelines | Submit actual work; observe completion and resulting artifacts. |
Expected results come from requirements, documented contracts, known bugs, or an independent reference. The program's own output does not become the oracle just because it was produced.
Compilation, startup, and passing unit tests provide evidence. Completion also requires observing the behavior the change promises. Existing automated checks are run when required or useful; generated exploration remains bounded and records its seed, configuration, and actual case count.
A missing dependency or inaccessible interface is reported as BLOCKED. The skill uses the permissions already granted to the agent and calls for isolated test data or sandboxes when operations have external effects.
The full instructions are in SKILL.md.
MIT. Copyright (c) 2026 Corvidae-Coding-Projects.