Smoke Tests

Run the project's smoke test suite, handling any build or test failures.

Arguments

The skill accepts optional arguments: a worker count for parallel execution, a build mode (debug or release), and/or a test filter (e.g., /smoke-tests, /smoke-tests workers=9, /smoke-tests workers=9 test_replication, /smoke-tests debug test_replication).

Ports for loreserver are allocated dynamically by the test harness (free ephemeral ports asked from the OS), so concurrent invocations of this skill don't collide and there's no per-session port bookkeeping to do.

Workers (optional, workers=N)

An argument matching workers=N enables pytest-xdist parallel execution with N workers by adding -n N to the pytest command. If omitted, no -n flag is added and tests run sequentially.

All workers in a run share a single loreserver launched by gw0; secondary worker fixtures used in test_replicated_store and test_topology allocate their own free ephemeral ports dynamically. Pick whatever worker count fits the host — there's no per-session cap.

Build Mode (optional, debug or release)

Pass the literal token debug or release to set the Cargo build profile. If omitted, the mode is release, for a filtered run and a full run alike, as docs/developing/code-standards/testing.md describes in Build profile: the release profile compiles incrementally, and one build serves every test run.

Test Filter (optional)

Any argument that isn't workers=N or debug / release is treated as a test filter and passed to pytest via -k. This uses pytest's keyword expression matching, which supports:

• Module name: test_replication — runs all tests in files/classes matching that name
• Test function name: test_basic_put — runs tests matching that name
• Compound expressions: test_replication and test_basic — combines filters with and, or, not

Examples:

• /smoke-tests — all smoke tests, sequential, release build (no filter)
• /smoke-tests workers=9 — all smoke tests, 9 parallel workers, release build (no filter)
• /smoke-tests test_replication — replication tests, sequential, release build
• /smoke-tests workers=4 test_replication — replication tests, 4 workers, release build
• /smoke-tests test_basic_put — test_basic_put, release build
• /smoke-tests workers=9 "test_replication and not test_slow" — compound filter, 9 workers, release build
• /smoke-tests debug test_replication — replication tests with debug build (override; rarely useful)

Instructions

Step 1: Parse Arguments

Split the provided arguments on whitespace:

1. Arguments matching workers=N set the xdist worker count. If none is given, no -n flag is added (sequential execution).
2. Arguments matching debug or release set the build mode explicitly.
3. All other arguments are treated as a pytest -k test filter. If none is given, no -k flag is added (all smoke tests run).

Then resolve BUILD_MODE:

• If debug or release was passed explicitly, use it.
• Otherwise, use release.

Step 2: Build the Project

Run the Cargo build for the resolved BUILD_MODE:

# release
cargo build --release -p lore-client -p lore-server --bin lore --bin loreserver

# debug
cargo build -p lore-client -p lore-server --bin lore --bin loreserver

If the build fails:

• Analyze the compiler errors
• Use EnterPlanMode to propose fixes
• After plan approval, implement the fixes and rebuild

Step 3: Run Smoke Tests

Once the build succeeds, run the smoke tests. Optionally append -n ${WORKERS} if workers was specified, and -k "${TEST_FILTER}" if a test filter was provided. Do not pass --lore-remote-*-port or --lore-remote-url — the harness allocates ports dynamically.

Use the binaries from ./target/${BUILD_MODE}/ (i.e. ./target/release/ or ./target/debug/).

Base command (sequential, all smoke tests):

uv run pytest scripts/test -m=smoke --log-cli-level=INFO \
  --lore-server-binary=./target/${BUILD_MODE}/loreserver \
  --lore-client-binary=./target/${BUILD_MODE}/lore

With workers (parallel execution): append -n ${WORKERS}

With test filter: append -k "${TEST_FILTER}"

Example: no filter, no workers:

uv run pytest scripts/test -m=smoke --log-cli-level=INFO \
  --lore-server-binary=./target/release/loreserver \
  --lore-client-binary=./target/release/lore

Example: 4 workers, filter test_replication:

uv run pytest scripts/test -m=smoke --log-cli-level=INFO \
  --lore-server-binary=./target/release/loreserver \
  --lore-client-binary=./target/release/lore \
  -n 4 \
  -k "test_replication"

Step 4: Handle Failures

If either build or tests fail:

1. Analyze the error output to understand the root cause
2. Use EnterPlanMode to propose a fix plan
3. After user approval, implement the fixes
4. Re-run the failed step to verify the fix
5. Continue until all tests pass

Notes

• Always build before running tests to ensure binaries are up-to-date
• The smoke tests require both loreserver and lore binaries from ./target/${BUILD_MODE}/
• The default build mode is release, with or without a filter, as testing.md describes. Pass debug to override it.
• Tests marked swfs need the EpicVFS driver (Windows only). If the machine lacks the driver, use -m "smoke and not swfs" in place of -m=smoke.
• Read relevant code standards from docs/developing/code-standards/ when fixing issues
• Use a 20-minute timeout for both build and test commands as they can take a while
• Do not use head or tail to view build or test output. Instead, redirect output to a timestamped temp file (e.g., cargo build --release -p lore-client -p lore-server --bin lore --bin loreserver 2>&1 | tee /tmp/smoke-build-20260303-1425.log) and use the Read tool to inspect the results. The timestamp disambiguates concurrent runs.