.PHONY: lint lint-shell test-shell format format-check typecheck test test-ci test-ci-serial test-ci-forked test-integration imports-check fuzz-smoke engine-archs docs-settings docs-formats check clean install demo demo-prep demo-publish build publish release promote release-promote docs docs-api docs-site site site-serve site-tar dns-setup qa-pod-volume qa-pod-up qa-pod-logs qa-pod-down

lint: lint-shell

uv run ruff check src/ tests/ tools/qa/ scripts/qa/ tools/readme_media.py tools/gen_settings_reference.py tools/gen_formats_table.py hatch_build.py

uv run python scripts/check_style_rules.py

# docs/settings.md is generated from the Config fields; a new setting must

# land in the reference with the code that adds it.

uv run python tools/gen_settings_reference.py --check

# The README formats table is generated from discovery's format map, so an

# xberg or tree-sitter bump that changes the set lands in the table.

uv run python tools/gen_formats_table.py --check

lint-shell: ## shellcheck the release scripts and actionlint every workflow

@if command -v shellcheck >/dev/null; then \

shellcheck -s bash scripts/*.sh tools/ci/*.sh tools/qa/artifact_smoke.sh tests/shell/*.bash tests/shell/stubs/*; \

else echo "lint-shell: shellcheck is not installed, skipping"; fi

# actionlint also shellchecks every inline `run:` block.

@if command -v actionlint >/dev/null; then actionlint; \

else echo "lint-shell: actionlint is not installed, skipping"; fi

test-shell: ## Run the bats tests for the release scripts

@command -v bats >/dev/null || { echo "test-shell: bats-core is missing (brew install bats-core)"; exit 1; }

bats tests/shell/

format:

uv run ruff format src/ tests/ tools/qa/ scripts/qa/ tools/readme_media.py tools/gen_settings_reference.py tools/gen_formats_table.py hatch_build.py

format-check:

uv run ruff format --check src/ tests/ tools/qa/ scripts/qa/ tools/readme_media.py tools/gen_settings_reference.py tools/gen_formats_table.py hatch_build.py

typecheck:

uv run mypy src/lilbee/

# The GPU QA tools read lilbee's own types, so they rot silently when one

# changes: multi_gpu_smoke crashed on its first check from #540 until this

# was added. The rest of tools/qa/ is not clean yet and is not listed.

uv run mypy tools/qa/placement_matrix/ tools/qa/multi_gpu_smoke.py

# hatch_build.py builds the PyPI long description, so a break here breaks

# the release rather than a test.

uv run mypy tools/readme_media.py tools/gen_formats_table.py hatch_build.py

test:

uv run pytest --cov=lilbee --cov-report=term-missing -v -n logical --dist loadgroup

test-ci:

uv run pytest --cov=lilbee --cov-report=term-missing --cov-report=html $(COV_FLAGS) -v -n logical --dist loadgroup

test-ci-serial:

uv run pytest --cov=lilbee --cov-report=term-missing --cov-report=html -v -p no:xdist

test-ci-forked:

uv run pytest --forked -v -n logical --dist loadgroup

imports-check:

uv run python -c "import lilbee; from lilbee import cli; from lilbee.core import config; from lilbee.data.extract import chunk, code_chunker; from lilbee.data import store, ingest; from lilbee.retrieval import embedder, query"

# The hosted-key tests need the litellm extra, which only this job installs.

test-integration:

uv run pytest tests/integration/ tests/providers/test_hosted_request_key.py -v

fuzz-smoke: ## Seeded adversarial TUI fuzz, fixed seeds (deterministic, CI-sized)

uv run python scripts/qa/tui_fuzz.py smoke

check: lint format-check typecheck test test-shell ## Run all checks (same as CI)

install:

uv tool install ".[crawler]" --force --reinstall --compile-bytecode

crawl-setup: ## Download Playwright Chromium for /crawl

uv run playwright install chromium

dns-setup: ## One-time DNS setup for lilbee.sh at Porkbun (reads creds from pass)

bash scripts/porkbun-dns-setup.sh

demo-prep: ## Stage demo data dirs via the gh-pages worktree

bash scripts/demo.sh prep

demo: ## Render every demo GIF via the gh-pages worktree

bash scripts/demo.sh render

demo-publish: ## Commit + push refreshed renders on gh-pages

bash scripts/demo.sh publish

build:

uv build

publish: build ## Build and upload to PyPI

uv publish

release: ## Bump the beta version, tag, and push; CI builds + publishes

bash scripts/release.sh

promote: ## Re-release an existing tag's source under a new version (FROM=v0.6.90b420.dev719 TO=0.6.90b420; DRY_RUN=1 to preview)

@[ -n "$(FROM)" ] && [ -n "$(TO)" ] || { echo "promote: FROM=v... TO=<new-version> required"; exit 1; }

bash scripts/promote_release.sh $(FROM) $(TO)

release-promote: ## Manual promote: rewrite notes as headings and mark a release latest (TAG=... or newest). verify-release.yml promotes on its own; use this to rewrite notes or to promote a tag it never ran for

@tag="$(TAG)"; \

[ -n "$$tag" ] || tag=$$(gh release list --repo tobocop2/lilbee --limit 30 --json tagName -q "first(.[].tagName | select(startswith(\"v\")))"); \

verified=$$(gh run list --repo tobocop2/lilbee --workflow=verify-release.yml --limit 50 --json displayTitle,conclusion -q "[.[] | select(.displayTitle == \"Verify release $$tag\" and .conclusion == \"success\")] | length"); \

if [ "$$verified" = "0" ]; then \

echo "REFUSING to promote $$tag: no green 'Verify release' run for it."; \

echo "The verify-release workflow must pass against the release assets first:"; \

echo " gh workflow run verify-release.yml -f tag=$$tag"; \

exit 1; \

fi; \

if gh release view "$$tag" --repo tobocop2/lilbee --json body -q .body | head -1 | grep -q '^Promoted from'; then \

echo "release-promote: $$tag (verified; promoted release, keeping its copied notes)"; \

gh release edit "$$tag" --repo tobocop2/lilbee --prerelease=false --latest; \

else \

prev=$$(gh release list --repo tobocop2/lilbee --exclude-drafts --limit 30 --json tagName -q "first(.[].tagName | select(startswith(\"v\") and . != \"$$tag\"))"); \

echo "release-promote: $$tag (verified; notes diff from $$prev)"; \

notes=$$(mktemp); \

bash scripts/release_notes.sh tobocop2/lilbee "$$tag" "$$prev" > "$$notes"; \

gh release edit "$$tag" --repo tobocop2/lilbee --notes-file "$$notes" --prerelease=false --latest; \

rm -f "$$notes"; \

fi

engine-archs: ## Regenerate the supported-architecture list from the pinned engine

uv run python tools/gen_supported_archs.py

docs-settings: ## Regenerate the cross-surface settings reference

uv run python tools/gen_settings_reference.py

docs-formats: ## Regenerate the README supported-formats table

uv run python tools/gen_formats_table.py

docs-api: ## Generate OpenAPI schema and Redoc static HTML

uv run python tools/gen_openapi.py openapi.json

npx --yes @redocly/cli build-docs openapi.json -o site/api/index.html

rm -f openapi.json

docs-site: docs-api ## Build the full dev portal (coverage + API docs)

# Tests gate CI and releases, not the site build; the coverage report deploys regardless of the test outcome.

-$(MAKE) test-ci

cp -r htmlcov site/coverage

uv run python tools/update_site_coverage.py

docs: docs-site ## Alias for docs-site

site: docs-api ## Build the deployable site locally (marketing pages + API docs; no coverage build)

@echo "site/ is ready. Preview with: make site-serve"

site-serve: site ## Build + serve the site at http://localhost:8000

cd site && python3 -m http.server 8000

site-tar: site ## Build the site and pack it into site.tar.gz

tar -czf site.tar.gz -C site .

@echo "wrote site.tar.gz"

# --- opencode QA pod (SkyPilot + RunPod) ------------------------------------

# Provisioning is SkyPilot; it reads the RunPod key from ~/.runpod/config.toml.

# One-time: pip install "skypilot[runpod]" && runpod config && sky check runpod.

# See tools/qa/opencode/README.md.

QA_SKY := tools/qa/opencode

qa-pod-volume: ## Create/adopt the reusable QA network volume (once)

sky volumes apply $(QA_SKY)/qa-volume.sky.yaml

qa-pod-up: ## Provision the QA pod, bootstrap, and run the matrix (MATRIX_ARGS=... to narrow)

# -i 30 --down: autostop after 30 min idle and tear the pod down when the matrix

# finishes, so a finished or hung run never bills idle. The volume keeps the

# engine, models, and results; reels just re-launch (a warm volume boots fast).

# HF_TOKEN (from your shell) is passed as a --secret so gated pulls work and

# the token stays redacted in sky logs; unset = ungated models only.

sky launch -c lilbee-qa $(QA_SKY)/qa-pod.sky.yaml -y -i 30 --down $(if $(HF_TOKEN),--secret HF_TOKEN="$(HF_TOKEN)",) $(if $(MATRIX_ARGS),--env MATRIX_ARGS="$(MATRIX_ARGS)",)

qa-pod-logs: ## Follow the QA pod's run log

sky logs lilbee-qa

qa-pod-down: ## Tear down the QA pod (the volume and its state survive)

sky down lilbee-qa -y

clean:

rm -rf .mypy_cache .pytest_cache .ruff_cache htmlcov .coverage dist/ openapi.json site/api site/coverage site.tar.gz

find . -type d -name __pycache__ -exec rm -rf {} +