diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 8e6c6f78..134342b3 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -39,7 +39,7 @@ jobs: run: uv sync --locked --all-extras --dev - name: Install tau2 for testing - run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@main + run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@v0.2.0 - name: Ruff lint run: uv run ruff check . @@ -82,7 +82,7 @@ jobs: run: uv sync --locked --all-extras --dev - name: Install tau2 for testing - run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@main + run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@v0.2.0 - name: Run Core Tests with pytest-xdist (Shard ${{ matrix.shard }}/${{ matrix.total-shards }}) env: @@ -121,7 +121,7 @@ jobs: run: uv sync --locked --all-extras --dev - name: Install tau2 for testing - run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@main + run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@v0.2.0 - name: Run Batch Evaluation Tests env: @@ -154,4 +154,4 @@ jobs: run: uv sync --locked --all-extras --dev - name: Install tau2 for testing - run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@main + run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@v0.2.0 diff --git a/.github/workflows/e2e-smoke-test.yml b/.github/workflows/e2e-smoke-test.yml index 33876979..b65f7d5a 100644 --- a/.github/workflows/e2e-smoke-test.yml +++ b/.github/workflows/e2e-smoke-test.yml @@ -37,7 +37,7 @@ jobs: run: uv sync --locked --all-extras --dev - name: Install tau2 for testing - run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@main + run: uv pip install git+https://github.com/sierra-research/tau2-bench.git@v0.2.0 - name: Run E2E Smoke Test id: run_test diff --git a/pyproject.toml b/pyproject.toml index 3025582f..e3d162a7 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -184,7 +184,7 @@ parentdir_prefix = "eval-protocol-" ### Black and isort removed; Ruff provides formatting and imports [tool.uv.sources] -tau2 = { git = "https://github.com/sierra-research/tau2-bench.git" } +tau2 = { git = "https://github.com/sierra-research/tau2-bench.git", rev = "v0.2.0" } [dependency-groups] dev = [