Configuration¶

ci-test-gate has no config file. Every knob is a CLI flag, and the two that also read the environment are the LLM settings.

Config file changes (`--config-changes`)¶

The one flag with real consequences for how many tests you run. Full output examples are in Usage → Config file changes.

Value	Behaviour
`broad` (default)	When a recognised project/dependency config file changes, recommend every test you supplied. Directly matched tests stay `required`, the rest become `recommended`.
`normal`	Ordinary path-based matching, config files treated like any other file.

ci-test-gate suggest --diff pr.diff --test-files tests.txt --config-changes broad

The recognised basenames are listed in Usage → Config file changes. They are matched on basename, so services/api/package.json and monorepo/crates/inner/Cargo.toml both count.

LLM mode¶

Pass --llm to suggest or local to classify with an OpenAI-compatible model instead of the built-in heuristic.

ci-test-gate suggest --diff pr.diff --test-files tests.txt --llm

Setting	Flag	Environment variable	Default
Enable	`--llm`	—	off
API key	`--llm-api-key`	`OPENAI_API_KEY`	—
Model	`--llm-model`	`CI_TEST_GATE_MODEL`	`gpt-4o-mini`
Base URL	—	`OPENAI_BASE_URL`	`https://api.openai.com/v1`

CLI flags take precedence over environment variables. There is no flag for the base URL — set OPENAI_BASE_URL to target a local or self-hosted OpenAI-compatible endpoint.

export OPENAI_API_KEY="sk-..."
export OPENAI_BASE_URL="http://localhost:11434/v1"

ci-test-gate suggest \
  --diff pr.diff \
  --test-files tests.txt \
  --llm \
  --llm-model gpt-4o \
  --output json

Fallback behaviour¶

If no API key is configured, or the LLM call fails for any reason — network error, rate limit, malformed response — ci-test-gate falls back to the heuristic classifier rather than failing. Your CI pipeline is never blocked by the LLM path.

$ ci-test-gate suggest --diff pr.diff --test-files tests.txt --llm   # no API key
## ci-test-gate Recommendation

**Reasoning:** Heuristic match: 1 required test(s) from 1 changed file(s).
...

The --config-changes setting applies in --llm mode as well.

What is not configurable¶

Test naming patterns. The heuristic reduces each test path to a base name by stripping test_, _test. and tests/ and then substring-matching it against the changed paths. There is no way to add or override a pattern.
The savings estimate. estimated_savings_pct is computed from test counts, capped at 95%. It is not configurable and not measured.
Classifier selection by name. TestClassifier.from_config() accepts classifier_type of heuristic or llm as a Python API detail; the CLI exposes it only as the presence or absence of --llm.

Language detection¶

The primary language of a diff is inferred from file extensions (src/ci_test_gate/context.py), and it feeds the LLM context only. The heuristic path is name-based and language-agnostic. Recognised extensions:

Extension	Language
`.py`	Python
`.js`, `.jsx`	JavaScript
`.ts`, `.tsx`	TypeScript
`.rs`	Rust
`.go`	Go
`.java`	Java

Anything else is unknown. See Language Support for the test naming conventions the classifier recognises.