test: wt test runs every suite, not just generator/

.config/wt.toml

@@ -32,6 +32,7 @@ url = "http://localhost:{{ branch | hash_port }}"
 [[pre-remove]]
 server = "lsof -ti :{{ branch | hash_port }} -sTCP:LISTEN | xargs kill 2>/dev/null || true"
 
-# The Python package lives under generator/; this alias runs pytest from the right cwd.
+# Tests live in four places (generator/, proxy/, the install-tend scripts,
+# worker/); dev/test.sh runs the lot, mirroring ci.yaml's test jobs.
 [aliases]
-test = "cd generator && uv run pytest {{ args }}"
+test = "dev/test.sh {{ args }}"

.pre-commit-config.yaml

@@ -29,15 +29,16 @@ repos:
     hooks:
       - id: actionlint
   # The composite actions' shared step bodies live in standalone scripts
-  # (shared/steps/, proxy/) rather than inline `run:` blocks, so
-  # actionlint's built-in shellcheck doesn't see them. Run shellcheck directly
-  # to keep that coverage. -S warning matches actionlint's default severity.
+  # (shared/steps/, proxy/, generator/src/tend/templates/) rather than inline
+  # `run:` blocks, so actionlint's built-in shellcheck doesn't see them. Run
+  # shellcheck directly to keep that coverage — and over dev/, which actionlint
+  # never sees at all. -S warning matches actionlint's default severity.
   - repo: https://github.com/shellcheck-py/shellcheck-py
     rev: v0.11.0.1
     hooks:
       - id: shellcheck
         args: ["-S", "warning"]
-        files: ^(shared/steps/.*\.sh|generator/src/tend/templates/.*\.sh|proxy/setup-sandbox\.sh)$
+        files: ^(shared/steps/.*\.sh|generator/src/tend/templates/.*\.sh|proxy/setup-sandbox\.sh|dev/.*\.sh)$
   - repo: local
     hooks:
       # The Claude Code slash-command preprocessor treats a backticked token

CLAUDE.md

@@ -11,13 +11,16 @@ completely — old formats should fail with a clear error, not silently parse.
 ## Commands
 
 ```bash
-wt test                            # run pytest in generator/
+wt test                            # every suite (generator/, proxy/, install-tend scripts, worker/)
 uvx tend@latest init               # regenerate workflows from .config/tend.yaml
 uvx tend@latest init --dry-run     # preview without writing
 uvx tend@latest check              # verify branch protection, secrets, bot access
-pre-commit run --all-files         # lint: ruff, typos, actionlint, uv-lock
+pre-commit run --all-files         # lint: ruff, typos, actionlint, shellcheck, uv-lock
 ```
 
+`wt test` runs [`dev/test.sh`](dev/test.sh), which mirrors the test jobs in
+`ci.yaml`; arguments go to the pytest suites (`wt test -k render`).
+
 ## Architecture
 
 Four pieces:
@@ -106,8 +109,8 @@ from `.config/tend.yaml`. Edit the generator or config, not the workflow files
 directly.
 
 The generator is a Python package under `generator/` — uses the uv_build
-backend, requires Python 3.11+. Runtime dependencies: click, ruamel.yaml.
-Dev dependencies: pytest, pytest-regtest.
+backend, requires Python 3.11+. Runtime dependencies: click, jinja2,
+ruamel.yaml. Dev dependencies: pytest, pytest-regtest.
 
 Consuming repos regenerate their `tend-*.yaml` workflows nightly (tend itself
 included — it dogfoods its own workflows). Changes to the generator do not

dev/test.sh

@@ -0,0 +1,65 @@
+#!/usr/bin/env bash
+# Every test suite in the repo, mirroring the test jobs in
+# .github/workflows/ci.yaml. `wt test` runs this (see .config/wt.toml), so one
+# command covers generator/, proxy/, the install-tend scripts, and worker/.
+#
+# Arguments are forwarded to the pytest suites (`wt test -k render`); a filtered
+# run skips worker/, whose vitest CLI takes different flags. Every suite runs
+# even if an earlier one fails, and the failures are listed at the end.
+set -o pipefail
+
+cd "$(dirname "${BASH_SOURCE[0]}")/.." || exit 1
+
+failed=()
+
+# suite <dir> <cmd>...
+suite() {
+  local dir=$1 rc=0
+  shift
+  printf '\n==> %s: %s\n' "$dir" "$*"
+  (cd "$dir" && "$@") || rc=$?
+  # A filtered run gets slack: pytest exits 5 for "no tests collected" and 4 for
+  # a path that only exists in another suite. generator/ stays strict, so a bad
+  # flag — 4 everywhere — still fails the run.
+  if [ ${#pytest_args[@]} -gt 0 ] &&
+    { [ "$rc" -eq 5 ] || { [ "$rc" -eq 4 ] && [ "$dir" != generator ]; }; }; then
+    rc=0
+  fi
+  if [ "$rc" -ne 0 ]; then failed+=("$dir"); fi
+}
+
+pytest_args=("$@")
+
+suite generator uv run pytest "${pytest_args[@]}"
+
+# The proxy addon isn't part of the generator package, and it imports
+# mitmproxy.test, so it runs standalone against the version production runs
+# rather than whatever mitmproxy is latest.
+if mitmproxy_version=$(yq -e '.inputs.mitmproxy_version.default' claude/action.yaml); then
+  suite proxy uv run --no-project --with pytest \
+    --with "mitmproxy==$mitmproxy_version" pytest "${pytest_args[@]}"
+else
+  echo "==> proxy: cannot read mitmproxy_version from claude/action.yaml (yq installed?)" >&2
+  failed+=(proxy)
+fi
+
+suite plugins/install-tend/skills/install-tend/scripts \
+  uv run --no-project --with pytest pytest "${pytest_args[@]}"
+
+if [ ${#pytest_args[@]} -eq 0 ]; then
+  # Install when the tree is missing or older than the lockfile (`npm ci` writes
+  # node_modules/.package-lock.json), and with `npm ci` rather than `npm install`
+  # — an older local npm reruns resolution and rewrites package-lock.json,
+  # leaving churn in the diff that has nothing to do with the change under test.
+  if [ ! -d worker/node_modules ] ||
+    [ worker/package-lock.json -nt worker/node_modules/.package-lock.json ]; then
+    suite worker npm ci --prefer-offline --no-audit --no-fund
+  fi
+  suite worker npm run typecheck
+  suite worker npm test
+fi
+
+if [ ${#failed[@]} -gt 0 ]; then
+  printf '\nfailed: %s\n' "${failed[*]}" >&2
+  exit 1
+fi