Agent Skills

BigTesty ships four Agent Skills: when a developer asks a coding agent to test BigQuery SQL or a pipeline, the agent writes the bigtesty.yaml, the test files or the Python tests the BigTesty way, instead of guessing them.

Skill	Use it to
setup	install BigTesty, authenticate, write the bigtesty.yaml (labels, CMEK…), run the tests in CI, upgrade from 0.x
writing-tests	write the test files: given / when / then, assertions, typed comparison, --update
python-tests	test Python code, dbt or another language on an ephemeral infra: BigTestyInfra, bigtesty infra
troubleshooting	go from a failed run (assertion diff, exit code 2, BigQuery error) to a cause and a fix

The agent loads a skill on its own when the conversation calls for it.

Install

Two ways, same skills.

From the CLI, the skills of the BigTesty version of the project:

bigtesty skills install                  # this project: .claude/skills and .agents/skills
bigtesty skills install --user           # the home directory: ~/.claude/skills and ~/.agents/skills
bigtesty skills install --target claude  # claude | agents | all (default)

.claude/skills is read by Claude Code, .agents/skills by the other agents following the Agent Skills convention. The copies are named bigtesty-<skill> and only replace the BigTesty folders. Committed in a project, they give the whole team the skills matching the BigTesty version locked in uv.lock: after an upgrade, bigtesty says that they are stale and bigtesty skills install refreshes them.

As a Claude Code plugin, kept up to date from the repository:

/plugin install bigtesty --marketplace tosun-si/bigtesty

The plugin version follows the BigTesty version. To call a skill explicitly: /bigtesty:writing-tests with the plugin, /bigtesty-writing-tests with the copies.

The skills live in the repository, in src/bigtesty/plugin/skills: they change with the code, in the same pull requests.