The five rows #2323 left as "assess", each either stated with its
reasoning and measurement, or shared with a constant that already has one.
- F4/F5: `driveAppFlow` and `connectViaDeepLink` default to
`BROWSER_TIMEOUTS.ui` / `.roundTrip` instead of restating 30_000 /
45_000 by hand; values unchanged, now one place to raise. A test in each
helper pins the defaults.
- F1: `announced-child.mjs` states `DEFAULTS` ({ timeoutMs: 30_000,
pollMs: 250 }) with the measurement — boot-to-announce of the composable
test server is 135ms median at load average 15, and a passing start pays
one poll interval, not the budget.
- F7: `pty.mjs` names `SCRIPT_PROBE_TIMEOUT_MS` (5s; the probe spawns in
1.3ms median, 3.6ms max under load) and stops folding ETIMEDOUT into
"no script(1)": a starved probe now fails `smoke:tui` loudly instead of
turning it into an exit-0 skip.
- G1: every job in `.github/workflows/main.yml` declares
`timeout-minutes` as a hung-job guard — roughly twice the slowest run
observed over the last 40 push runs (and the last three releases for the
publish jobs), rounded up to the next five minutes, with the range stated
per job.
AGENTS.md's timeout section now points at where these live instead of at
this issue.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PsmUAZ6aLUAFqenjJKSSRp
Signed-off-by: cliffhall <cliff@futurescale.com>
Copilot review round 5 — four findings, all in the suppressed block.
1+4. The regression stubs did `console.log(MARKER); process.exit(1)`.
Writing to a PIPE is asynchronous and `process.exit()` tears the
process down without draining, so the marker can be lost outright —
and then the stub that pins the paint-then-die path fails as "exited
before rendering", the neighbouring branch and the wrong diagnostic
for the thing it guards.
Worth recording that I met this mechanism last round and did not join
the dots: probing the drain race, a 4MB write followed by
`process.exit()` reached the parent with the marker missing
entirely. That was noted as a reason the repro failed, while the same
construct stayed in the stubs. Both now set `process.exitCode` and
let the loop drain — still exiting immediately after the write — with
a comment, since `process.exit()` is the more obvious spelling and
will read as a simplification to the next person. Same fix for the
`console.error("boom")` stub, whose assertion had the same exposure.
2. DEFAULTS claimed "every one is a hard floor as well as a default",
which contradicts `normalizeMs` (any finite positive value passes)
and the suite two files over, which passes 400ms against a 2000ms
default. It rejects the shape a bad env var produces, not smallness.
3. The raw-mode forbidden-output check was unreachable from any test:
the real PTY happy path only exercises the pattern NOT matching, so
a regression breaking it would stay green forever — an
acceptance-criterion guard with nothing guarding it, which is the
recursive form of this PR's own complaint.
Now a `runRenderSmoke` option (`forbidOutput`), so it runs inside the
tested state machine and against the fully drained output.
`smoke:tui` passes its pattern in and drops the post-hoc check. It
can only turn a pass into a failure, never rewrite an existing
failure's message: a crash reason is more useful than "and it also
printed X", which is usually a symptom of that crash. Three tests —
matching, not matching, and that asymmetry.
Verified: test:scripts 213/213; `smoke:tui OK — rendered "MCP Servers" at
360ms and was still running 2000ms later`; and the no-PTY shim still
fails through the refactored path.
Signed-off-by: Cliff Hall <cliff@futurescale.com>
Signed-off-by: cliffhall <cliff@futurescale.com>
Two findings, both of which quietly undo the PR.
1. `SMOKE_TUI_SURVIVE_MS` was unvalidated. `??` guards only nullish, so
an env var set to the empty string reaches `Number("")` -> 0 and a typo
reaches NaN; `setTimeout` fires immediately on both. The survival check
would then resolve before the child could possibly have died, printing
"still running 0ms later" and asserting nothing — the exact first-paint
false green this PR removes, restored by an empty environment variable,
on a green run.
Fixed at the module rather than only at the env-parse site, so every
caller is covered: `normalizeMs` (same shape as child-cleanup.mjs's
`normalizeGraceMs`, matching an existing policy rather than inventing a
second one) applied to all four durations before anything reads them,
which also stops the messages interpolating "NaNms". Defaults moved to
an exported DEFAULTS so the smoke and the module cannot drift.
`surviveMs: 0` is deliberately NOT honored as "skip the wait": no
caller wants that, and it is indistinguishable from the typo.
2. (suppressed comment) `spawnSync` REPORTS ENOENT in `result.error` — it
does not throw — so `probeScriptVersion`'s try/catch read a missing
binary as "ran, printed nothing". On linux that empty output fell
through to the util-linux guess, and the smoke then hard-failed
spawning a binary that does not exist instead of taking the documented
skip.
`probeScriptVersion` now returns `{ available, output }`, and
`resolvePtyWrapper` returns `{ ok: true, flavor, wrap }` or
`{ ok: false, reason }` — the two unavailability causes deserve
different skip messages, since "no script(1) on PATH" and "nobody has
verified this platform's invocation" send the reader elsewhere. A
non-zero exit is explicitly not unavailability: BSD `script` has no
`--version` and answers with its usage, which is the evidence the
flavor fallback reads.
Tests for both, since each finding lives entirely in a branch no happy
path reaches: the normalizeMs table (empty string, typo, undefined, -1,
Infinity, a numeric string, 0), an end-to-end run proving a
`surviveMs: Number("")` paint-then-die stub still fails, the
`result.error` ENOENT and timeout shapes, and both `resolvePtyWrapper`
failure reasons.
Verified: `SMOKE_TUI_SURVIVE_MS="" npm run smoke:tui` reports "still
running 2000ms later"; with PATH blanked, resolvePtyWrapper returns
`{ok: false, reason: "no `script(1)` on PATH"}` rather than guessing.
Signed-off-by: Cliff Hall <cliff@futurescale.com>
Signed-off-by: cliffhall <cliff@futurescale.com>
`smoke:tui` spawned the launcher with `stdio: ["ignore", …]`, waited for
"MCP Servers" in the output, printed OK, and SIGTERMed. It never looked at
the child again.
The child was already dead. Ink mounts `useInput`, `useInput` needs raw
mode, and raw mode is a property of the file descriptor — with stdin on
/dev/null the TUI painted one frame and exited 1 about 40ms later with
"Raw mode is not supported on the current process.stdin". Measured here:
first frame at 396ms, exit 1 at 433ms. The smoke won that race and
reported success, so what it asserted was *first paint*, not survival —
contradicting its own header ("boots and renders without crashing").
This is not environment-specific: `stdio: ["ignore", …]` makes the child's
stdin /dev/null whether or not the developer's terminal is a TTY, so this
smoke has never verified a running TUI on any machine. And the race cut
both ways — an `exit` processed first failed the run, which reads as flake
rather than as the standing defect.
Two pieces:
- `scripts/lib/pty.mjs` gives the child a real terminal via `script(1)`,
no dependency. The three flavors are not interchangeable (BSD takes
argv; util-linux takes one shell string plus `-e`; busybox that string
without `-e`), so the invocation is built and unit-tested rather than
written inline. A platform with no `script(1)` skips loudly instead of
running without one, which would be a guaranteed failure rather than a
weaker check.
- `scripts/lib/render-smoke.mjs` is the fix: the child must still be
running `surviveMs` (default 2s) after first paint. Without that, a
future harness change reintroduces the same false green and nothing
notices — which is why it lives in a module `test:scripts` drives
against a stub that paints the marker and immediately exits. Today's
harness passes that stub; this one fails it, with a diagnostic that
distinguishes "died after painting" from "died before painting".
The smoke additionally fails if the raw-mode error appears in a passing
run, so a future Ink that degrades to a warning instead of throwing can't
ship a TUI with no keyboard input past this gate.
Verified end to end: with a `script` shim that allocates no pty, the smoke
now fails with "it painted one frame, it did not run"; with the real one
the TUI is still running 2s after its first frame and never logs the
raw-mode error.
The `process.env.CI` skip stays, but its comment is corrected: the PTY
removes the technical blocker it used to cite, so the smoke is local-only
by decision now, not by capability. Whether it joins GitHub CI is out of
scope here (#2146 is keeping the other local-only smokes out on purpose).
Signed-off-by: Cliff Hall <cliff@futurescale.com>
Signed-off-by: cliffhall <cliff@futurescale.com>
Five `scripts/*.mjs` consumers each carried their own `ensureTestServer()`
that returned early when `test-servers/build` was already on disk. Presence
is not freshness: once that directory exists — which it does on any machine
that has run the smokes once — an edit to `test-servers/src` was never
picked up again, and the smoke went on driving the previously-built fixture.
The failure is silent and misattributed. It does not read as "your fixture
is stale"; it reads as a product bug in whatever the smoke is checking.
`scripts/lib/ensure-test-servers.mjs` is now the one place that build is
produced for a script consumer, and it builds unconditionally — memoized
once per process per repo root, so pack-and-verify's two asks stay one tsc
pass while a fresh run always rebuilds. Callers name the entries they
depend on and get one actionable failure message, replacing five that had
drifted. pack-and-verify also stops shelling out to `npx tsc`, which is
what made it the one script unrunnable on Windows (#1939).
`test-servers/tsconfig.json` gains `incremental` so the up-to-date case is
cheap: 3.0s cold, 1.6s incremental here.
Unconditional emit is still not a clean — `--noCheck` leaves the stale .js
of a deleted src file behind — which is unchanged from before and recorded
in the helper's header and AGENTS.md.
Closes#2111
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JTHVxSu8AUgHRLvo1ntZ8H
Signed-off-by: cliffhall <cliff@futurescale.com>
On Windows npx/npm are .cmd shims a shell-free execFileSync/spawnSync
cannot start (ENOENT since the CVE-2024-27980 hardening), so
verify:typecheck-coverage died at validate's second step — doubly
silently, echoing "(no diagnostic captured)" per project and then
reporting all 918 tracked files as getting no tsc pass — and
verify:build-gate, the smoke test-server bootstraps, and pack:verify's
npm pack failed the same way.
Add scripts/lib/resolve-node-bin.mjs: resolves the JS entry behind a
package's bin from <pkg>/package.json (deep resolution is blocked by
Vite 8's exports map) and spawns it via process.execPath — the same
walk npx --no-install did, cross-platform and shell-free. Switch the
six npx call sites to it; the npm pack call gets the existing
shell-on-win32 idiom instead (npm has no in-tree package to resolve).
An unresolvable tsc is now a hard "cannot measure" error with
actionable stderr, pinned by a new main() regression test.
Closes#1939
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(scripts): await the pack:verify web child's exit before removing its work dir
`pack-and-verify.mjs` had the same kill-then-remove shape that made
`smoke:tui` flake with ENOTEMPTY in #1801: `verifyWeb()`'s
`finally { stop() }` only *delivered* SIGTERM, then the main flow
`rmSync`'d the work dir immediately, and `fail()` did the same back to
back.
Applies the #1814 treatment, factored into a shared
`scripts/lib/child-cleanup.mjs` so the two scripts can't drift:
- `stopChild()` — SIGTERM, await the child's `exit`/`close`, escalate to
SIGKILL after a grace period, and ultimately proceed with a warning so
a wedged child can never hang the script. `verifyWeb()`'s `stop()` now
goes through it and is awaited, so the child is gone before the caller
removes `work`.
- `removeSafe()` — `rmSync` that warns instead of throwing. Used on both
success-path removals (a leftover temp dir must never fail a run that
passed) and in `fail()`, where an ENOTEMPTY could otherwise bury the
real diagnostic under an rmSync stack and skip the "tarball retained
at ..." hint. `fail()` stays synchronous — it is called from sync
contexts and ends in `process.exit()` — so it best-effort signals the
child; `removeSafe()` is what makes that safe.
`smoke-tui.mjs` adopts `removeSafe()` for its `cleanup()`. Its `done()`
keeps its bespoke wait (documented inline): that one must keep waiting
for `close` *after* `exit` on a shorter re-armed deadline so the quoted
output is complete, which `stopChild()`'s first-of-exit/close contract
deliberately does not do.
Adds `scripts/lib/child-cleanup.test.mjs` (10 cases, `node --test` via
`npm run test:scripts`) covering the SIGTERM/SIGKILL escalation, the
`close`-only spawn-failure path, the never-hang give-up, late events
after giving up, and that `removeSafe` warns rather than throws.
`npm run ci` passes, and `npm run pack:verify` (not in the gate — needs
network) was run end to end and passes with no cleanup warnings.
Claude-Session: https://claude.ai/code/session_01YAt8rqxysNbhYWLhoRm3fU
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(scripts): normalize stopChild's graceMs and correct removeSafe's JSDoc
Copilot review round 1 on #1895.
- `stopChild()` trusted `graceMs` as given, but call sites build it as
`Number(process.env.X ?? 5000)` — a typo yields `NaN`, which
`setTimeout` treats as `0`. That would fire SIGTERM and SIGKILL in the
same tick and print "did not exit within NaNms", collapsing the whole
escalation silently. New exported `normalizeGraceMs()` falls back to
`DEFAULT_EXIT_GRACE_MS` for anything non-finite or non-positive, with
two tests: the table of bad values, and an end-to-end one asserting a
`NaN` grace period still leaves only SIGTERM sent a turn later.
- `removeSafe()`'s `@returns` said "true if the path was removed", which
is inaccurate for a missing path (`force: true` makes it a no-op
success). Reworded to say what it actually reports.
`npm run ci` passes.
Claude-Session: https://claude.ai/code/session_01YAt8rqxysNbhYWLhoRm3fU
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix: wait for the TUI to exit before removing its temp dir (#1801)
`done()` sent SIGTERM and then immediately `rmSync`'d the work dir, which
doubles as the TUI's HOME. Nothing waited for the child, so the Ink process
could still be writing into the dir while rmSync walked it — a file landing
after a directory was read but before it was removed surfaces as ENOTEMPTY
on macOS, failing `smoke:tui` after a successful render.
Removal now happens only once the child's `exit` event fires: SIGTERM, then
SIGKILL after SMOKE_TUI_EXIT_GRACE_MS (default 5s), then clean up regardless
after twice that. The crash-before-render path short-circuits via a
`childExited` flag set by the pre-registered exit handler. cleanup() also
warns instead of throwing, so a leftover temp dir can never fail a passing
smoke.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MvKF2mqi4hoQSUqQpZXNTq
* fix: wait on close, not exit, so spawn failure resolves instantly (review #1814)
A failed spawn emits `error` + `close` and never `exit`, so the previous
`child.once("exit")` wait never fired on that path: done() sent SIGTERM to a
process that never existed and resolved only via the give-up timer — 10s at
the default grace, with a warning blaming a SIGTERM the child never received.
Waiting on `close` covers both normal termination and spawn failure, and
additionally guarantees the stdio pipes are drained.
The short-circuit gate moves with it, `childExited` → `childClosed`: keying it
on `exit` would have skipped the drain on the crash-before-render path, which
is exactly where the quoted output matters. Messages that quote that output are
now thunks rendered in finish(), after the drain, instead of being sliced at
call time. SIGTERM is now sent only to a child that is still running, since on
both the crash and spawn-failure paths there is nothing left to signal.
finish() also gets a `finished` guard so its single-entry invariant is local
rather than inferred from process.exit() being synchronous.
Measured, spawn failure (grace shortened to 500ms): 1.04s + spurious warning
before, 0.02s and silent after. Crash-before-render now prints the child's
stderr detail line that the pre-drain slice could truncate. Happy path 3x,
timeout path, and full `npm run smoke` + `npm run validate` all green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MvKF2mqi4hoQSUqQpZXNTq
* fix: bound the drain, and make the give-up warning true (review #1814 r2)
Three follow-ons from the second review pass:
The give-up warning still said the TUI "did not exit within Xms of SIGTERM",
but after switching the wait to `close` neither clause is reliably true — the
timer's most plausible trigger is now a child that has already exited, and on
the crash and spawn-failure paths no SIGTERM is sent at all. That is the same
defect this PR fixed one round ago (a warning naming a signal never sent), so
the deadline is now re-armed with a message describing what is actually being
awaited.
`close` is bounded by whoever holds the stdio pipes, which can outlive the
direct child: a descendant that inherited them keeps it pending. That is the
one real cost of `exit` → `close`, and it is moot only because the TUI spawns
nothing at boot — a reason that lives entirely outside this file. Once `exit`
has fired the deadline drops to DRAIN_MS, keeping the drain guarantee without
inheriting an unrelated process's lifetime. Measured against a grandchild
holding the pipes for 3s: 0.54s and an accurate warning, instead of 3s.
The quoted output is sliced from the tail rather than the head. The thunk added
last round exists to capture bytes that arrive late, which a head slice then
discards as soon as output exceeds 800 chars — and on the timeout path the last
frame is what diagnoses a stuck render, not the first. Verified: a 2KB stderr
crash now reports its final line. The sibling scripts keep `slice(0, 800)`;
changing them is a deliberate sweep, not a drive-by here.
Also corrected the cleanup() comment — the wait lives in done(), and it is for
`close`, not `exit`.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MvKF2mqi4hoQSUqQpZXNTq
* docs: correct the cleanup() and DRAIN_MS comments; tail-slice cleanly (review #1814 r3)
The cleanup() comment claimed its warning "should not fire now that removal
waits for close" — true only of the normal path. Both give-up branches call
finish() without close, and those are precisely the states where something may
still hold the dir, so the warning is expected there. Same defect class as the
two warnings already fixed in this PR (a message asserting what the code no
longer guarantees), relocated into a comment.
The DRAIN_MS comment explained why the cap exists but not what it costs: on
that path the work dir is removed while a descendant holding HOME=work may
still be writing to it — #1801's race, re-entered deliberately. That is only
acceptable because cleanup() warns instead of throwing, so record the two as
load-bearing for each other where the constant is defined.
Tail-slicing an Ink stream can begin mid-CSI-sequence, leading a diagnostic
with an escape-code fragment that mangles the line after it. outputTail() drops
through the first newline when the output was actually truncated, and returns
short output untouched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MvKF2mqi4hoQSUqQpZXNTq
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Closes#1795.
Enforce localhost-only binding in code (the "never 0.0.0.0" rule was only a spec note): refuse an all-interfaces HOST — across every spelling the OS binds as the wildcard (inet_aton decimal/octal/hex, IPv6, IPv4-mapped, zone-scoped, IDNA/Unicode) — unless DANGEROUSLY_BIND_ALL_INTERFACES=true (Docker's sanctioned opt-in). Fixes the reported stdio-connect 403 by expanding the default loopback origin allow-list to all three interchangeable forms.
Hardening surfaced across an extended review: closes an IDNA/fullwidth wildcard bypass, a blank-ALLOWED_ORIGINS fail-open, and the CLI/TUI OAuth-callback wildcard bind (loopback-only now, RFC 8252); canonicalizes the origin allow-list the way a browser does; derives frame-ancestors from the real embedder list (IPv6-literal exclusion verified in headless Chromium); strictPort on all three bind points; and preserves the CLI exit-code map + JSON envelope through the mcp-inspector --cli launcher path. Shared host normalization lives in core/node/hostUrl.ts. Docs + tests + smokes throughout.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
* refactor(launcher): self-contained validate; split smokes into top-level step (#1504)
Previously `validate:launcher` built cli/tui/web and ran the cross-client
smokes, so the launcher's "validation" depended on every other client
being built — and the launcher had no lint/format/coverage of its own.
- clients/launcher: add a web-style `validate` (format:check && lint &&
build && test:coverage) with a Node-only eslint flat config, prettier,
and a per-file coverage gate (parse-launcher-argv.ts covered;
src/index.ts excluded as binary bootstrap). 100% on the gated file.
- Root `validate:launcher` now just delegates: `cd clients/launcher &&
npm run validate` — no dependency on the other clients.
- New top-level `smoke` step (smoke:launcher + smoke:cli + smoke:tui +
smoke:web) with NO build commands; `validate` runs it last. Each
client's own validation builds its bundle first, so nothing builds
twice (cli/tui via test:coverage, launcher via its validate;
smoke:web builds dist on demand).
- clients/tui test:coverage now builds the TUI bundle (mirroring cli) so
the build-free smoke has it available.
- CI: replace the standalone cli/tui/launcher build step with a
"Validate launcher" step. Docs updated (AGENTS.md, launcher README).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(cli,tui,root): every client self-validates; smoke separate from validate
Extends the launcher self-validate work to the CLI and TUI, and makes the
top-level `validate` nothing but the per-client validations — smokes are
now a separate top-level target.
- clients/cli, clients/tui: add a web-style `validate` with a Node flat
eslint config + prettier. CLI lint/format pass as-is. TUI registers
eslint-plugin-react-hooks for the classic rules only (rules-of-hooks +
exhaustive-deps, which the source already references); the stricter
react-hooks@7 rules are not enforced on the interim component surface
(#1501).
- Avoid double builds:
- CLI's e2e spawns the built binary, so its `test:coverage` builds first;
`validate` folds that in (`format:check && lint && test:coverage`, no
separate `build`).
- TUI tests don't spawn a bundle, so `validate` keeps an explicit
`build` (for the smoke) and `test:coverage` is pure.
- Root: `validate` = validate:web && validate:cli && validate:tui &&
validate:launcher (each delegates to the client's own validate). `smoke`
is no longer part of `validate` — it's a separate top-level target run
after a build/validate, with no build commands of its own.
- CI: per-client "Validate X" steps + a single "Run cross-client smokes"
(npm run smoke). Docs updated (AGENTS.md, cli/tui READMEs).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(root): drop redundant test aggregator scripts
Now that every client self-validates and the top-level `validate` chains
the per-client validations, the root `test` / `test:web` / `test:cli` /
`test:tui` / `test:coverage` aggregators are redundant — run
`npm run validate` (all) or `cd clients/<name> && npm run validate` (one).
`test:storybook` stays (it's the one check not part of validate). CI was
already using each client's own scripts, so it's unaffected. Docs updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(web): silence happy-dom infra noise in unit runs
Passing unit runs printed dozens of alarming-but-expected lines that made
green runs look broken — all from happy-dom performing real I/O with no
server behind it:
- `<iframe src="/sandbox.html">` (MCP Apps sandbox) made happy-dom fetch
and later unload that URL, emitting ~40 `DOMException [NetworkError/
AbortError]` + `AsyncTaskManager destroyed` lines. Disable child-frame
navigation in the unit project (the tests only assert on the iframe
element/attributes, not its document), removing the noise with no
coverage loss.
- App-level renders read `GET /api/config` via useSandboxUrl /
useServerListWritable, 404-ing ~76x. Add a benign default `fetch` (empty
200) in the unit setup; tests that exercise fetch still override it.
Negative tests that assert on errors already suppress their own console
output (e.g. jsonUtils/schemaUtils); CLI, TUI, and the web integration
suite were already clean. Unit suite is now noise-free; 2424 web tests
pass and the coverage gate is unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(scripts): make validate fast (test), split coverage into its own target
Each client's `validate` now runs `test` instead of `test:coverage`, so
the everyday pre-push check skips v8 instrumentation (and, for web, the
slower integration project). The per-file coverage gate moves to a
dedicated top-level `coverage` script:
coverage = coverage:web && coverage:cli && coverage:tui && coverage:launcher
coverage:<name> = cd clients/<name> && npm run test:coverage
`validate` still builds each client's bundle (cli via its test `pretest`,
the others via an explicit `build`), so `npm run validate && npm run
smoke` works unchanged. CI now runs `validate` (fast checks + builds),
then `coverage` (the gate), then `smoke` — collapsing the old per-client
steps. Docs updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* ci: drop the coverage gate from CI; keep web integration as its own step
CI no longer runs `npm run coverage` — the per-file coverage gate is now a
local/manual check (`npm run coverage` before pushing). To avoid silently
dropping the web integration suite (it only ran via test:coverage, and
`validate` runs the unit project only), CI gains a standalone
`clients/web` `test:integration` step that exercises the real
stdio/HTTP/OAuth/storage paths without coverage instrumentation.
CI is now: validate (fast) → web integration → smoke → storybook. Docs
updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(smoke): give smoke:launcher its own script like the others
smoke:launcher was the odd one out — an inline chain of three
`node clients/launcher/build/index.js ... --help` commands while smoke:cli
/ smoke:tui / smoke:web each run a `scripts/smoke-*.mjs`. Extract it to
`scripts/smoke-launcher.mjs`, which runs the same `--help` / `--cli --help`
/ `--tui --help` dispatch checks but now also asserts each prints that
mode's usage banner (proving the launcher resolved the right client build)
and verifies the required builds exist up front with a clear error. Prints
a `smoke:launcher OK — ...` summary in the same style as the rest. Docs
updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: reorder package.json scripts for readability
Move frequently-run scripts toward the top and rarely-run-by-hand ones
(pretest hooks, lint/format internals, etc.) toward the end, consistently
across the root and client manifests. Pure reordering — no script names or
values changed, and no other manifest content is touched.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: drop references to removed root inspector / test:storybook scripts
The reordering dropped the root `inspector` and `test:storybook`
aggregators (intentional — `inspector` is redundant with `web`, and
storybook is run from clients/web in CI). Update AGENTS.md so it no longer
points at `npm run inspector`, and run `test:storybook` from `clients/web/`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: address PR #1505 review (format glob gap, fetch-baseline comment)
- TUI format/format:check now also cover the client-root source files
(index.ts, tui.tsx, dev.ts) that live outside src/ — `eslint .` already
lints them, so they were linted but not format-checked. (cli/launcher
have no source outside src/; their root .ts are single-quoted configs
intentionally left out of prettier, matching web.)
- Reword the setup.ts default-fetch comment: it overstated auto-restore
(the unit project sets no restoreMocks/unstubGlobals), and now also
flags that the permissive 200 default means a test asserting on a fetch
FAILURE must install its own rejecting stub.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: remove redundant root install:clients script
`install:clients` ran the same `node scripts/install-clients.mjs` as
`postinstall`, which fires automatically on `npm install` — so the manual
alias was redundant. Drop it and the doc references (AGENTS.md, spec).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(ci): skip smoke:tui in CI (Ink needs a real TTY)
This PR made CI run the full `npm run smoke`, which surfaced that
smoke:tui fails in headless CI: the Ink TUI throws "Raw mode is not
supported" on mount (no TTY) and exits before rendering its first frame.
The old CI never ran smoke:tui, so this was latent.
smoke-tui.mjs now self-skips when `process.env.CI` is set — the boot/
render check is inherently local-only (a TUI needs a terminal). The TUI is
still built and unit-tested in CI; `npm run smoke` stays the single
command everywhere (full locally, tui-skipped in CI). Verified
`CI=true npm run smoke` is green (launcher/cli OK, tui SKIPPED, web OK).
Docs + CI comment updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(cli,tui): normalize --catalog / --config with web; update --help (#1347)
Bring the CLI and TUI in line with the web catalog vs read-only session
model introduced in #1481/#1483, and close the outstanding --help item
from #1347.
Core (core/mcp/node/config.ts):
- Add `catalogPath` to ServerConfigOptions (writable catalog slot).
- `resolveServerSource()` picks the active source: `--catalog` (writable,
seed-if-missing) wins over `--config` (read-only, error-if-missing).
- Writable sources seed an empty `{"mcpServers":{}}` on first run (matches
the web backend) instead of erroring; read-only `--config` still errors.
- `serverSourceConflict()` shared conflict matrix (mirrors run-web):
--catalog/--config mutually exclusive; neither combines with an ad-hoc
target.
- `readServerListFile()` public loader for runners (TUI) needing the raw map.
- Rename `withDefaultConfigPath` → `withDefaultCatalogPath`; the no-flag
default now fills the writable catalog slot, not --config.
CLI / TUI:
- Add `--catalog <path>` (+ MCP_CATALOG_PATH) and re-document `--config` as a
read-only session in --help.
- Enforce the shared conflict matrix before resolution.
Full writable persistence/CRUD threading remains deferred to #1482 / #1432
(CLI/TUI are read consumers today; the split surfaces as seed- vs
error-on-missing).
Tests: core resolution matrix (source/conflict/seed/read-only), CLI catalog
+ conflict integration tests, TUI loadTuiServers unit tests. READMEs updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(launcher): add smoke:cli + smoke:tui e2e launcher smokes (#1347)
The launcher smokes only checked `--help` for --cli/--tui; nothing
exercised the launcher → CLI/TUI → core path on the built artifact (only
smoke:web did, for web). Add two end-to-end smokes wired into
validate:launcher between smoke:launcher and smoke:web:
- smoke:cli (scripts/smoke-cli.mjs): runs `--cli` against the bundled
stdio test server via a temp --catalog and asserts tools/list returns
the tools, plus the #1347 resolution paths — default-catalog
seed-on-missing, read-only --config error-without-seed, and the
--catalog/--config conflict.
- smoke:tui (scripts/smoke-tui.mjs): launches `--tui --catalog <temp>`
and asserts the Ink app renders its first frame ("MCP Servers") within
a timeout, then SIGTERMs it (shallow boot/render check).
Both build test-servers/build on demand if missing. Docs updated
(AGENTS.md validate description, launcher README).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor: address review on #1499 (comment accuracy, CLI/TUI flag parity)
Review feedback from @claude on PR #1499 (all non-blocking):
- Align CLI catalogPath resolution with the TUI: `options.catalog?.trim()
|| process.env.MCP_CATALOG_PATH` so an explicit empty `--catalog ""`
falls back to MCP_CATALOG_PATH identically in both front-ends (CLI used
`??`, which only guards undefined/null).
- Clarify the `serverSourceConflict` comment: it mirrors only the
source-selection portion of web's matrix and deliberately omits web's
`--header` rejection, since CLI/TUI merge --header into per-server
settings.
- Note the deliberate --header broadcast semantics in loadTuiServers.
- Share an EMPTY_CATALOG_CONTENT constant between seedEmptyCatalog and
readServerListContent instead of two separate literals.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(cli): trim --config to match --catalog and the TUI (#1499 review)
Cosmetic uniformity follow-up from the re-review: configPath now uses
`options.config?.trim() || undefined`, matching the catalog slot above it
and the TUI. Behavior is unchanged (resolveServerSource/serverSourceConflict
already trim downstream).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
cliffhall
olaservo