Run end-to-end plugin-skill tests in real Claude subprocesses with TAP output.
npx skills add https://github.com/yoshisada/ai-repo-template --skill kiln-test
Please help me install this Agent Skill. Skill: kiln-test Source: https://github.com/yoshisada/ai-repo-template/tree/main/plugin-kiln/skills/kiln-test Command: npx skills add https://github.com/yoshisada/ai-repo-template --skill kiln-test
Runs executable tests against plugin skills — the real skill, in a real Claude subprocess, against a scratch-dir fixture, with assertions that verify final scratch-dir state.
Run /kiln:kiln-test to execute all tests for the detected plugin, or specify a plugin and test to run a single suite.
High-intent search queries and answers about installing and using this skill.
End-to-end tests for Claude plugins are automated by running real Claude subprocesses in a scratch directory. The testing harness executes plugin skills against fixture data and verifies final scratch-directory state through automated assertions.
Automated plugin testing verifies final scratch directory state by executing the real skill within a Claude subprocess against a fixture. After execution, the harness evaluates assertions against the resulting scratch directory contents to determine a pass or fail verdict.
TAP-based verdict reporting is a standardized output format used to communicate test results. The testing harness produces TAP v14 output and records verdicts under the .kiln/logs directory, providing a structured and readable summary of plugin test executions.
Yes, you can execute a single test suite instead of a full plugin run. The harness supports auto-detect plugin invocation, explicit full-plugin runs, and per-test execution forms, allowing you to target specific tests across plugin directories.
Yes, the test harness retains scratch directories for post-mortem analysis when tests fail. This allows developers to inspect the exact state of the fixture data and debug the Claude subprocess execution after a failing assertion.
A limitation of running real Claude subprocesses for plugin testing is the dependency on resolving the wheel-test-runner.sh script via dual-layout sibling traversal. The harness requires this specific directory structure to discover and execute the underlying test scripts successfully.
yoshisadaAutomate baseline-vs-candidate plugin research and emit a markdown report.
multicamAnalyze core dumps with gdb to diagnose early startup crashes and segfaults.
yoshisadaPrint a deprecation notice and delegate execution to kiln-next.
ruskibeatsManages background dispatch of Clanker Ops tasks via subprocess spawning and audit logging.
