fix: address Copilot review round 1 on #2248

Five findings, all real.

**Critical — `verifySkills` selected `contents[0]`.** These bytes are hashed
against the file's advertised digest, so accepting a block the server
labelled something else verifies one file's content against another file's
digest, and can report that as `verified`. A false pass is worse than a
missing check. `contentsFor(result, uri)` now selects by NORMALIZED identity,
which keeps the canonicalized-echo case that motivated the original code
while rejecting an unrelated block, and reports a read error when nothing
answers for the URI. This is what `onReadSkillFile` already did; the two
paths hash bytes against digests and must not disagree about which bytes.

**`skills/get` was dispatched without the extension gate** that `skills/list`
had. Hoisted into `assertSkillsSupported` so the two cannot drift — declaring
the extension commits a server to both. Without it, an undeclared server's
-32601 is indistinguishable to a script from the -32602 a declared server
returns for a URI it does not serve.

**The TUI could strand a user on a hidden Skills tab.** The tab left the bar
when the gate went false but `activeTab` did not, so the pane kept rendering
for a server that never declared the extension. Reset to `info`, following
the Auth precedent — additionally gated on `connected`, because this gate
reads a server declaration and would otherwise fire during a reconnect.

**The TUI keyed a verdict by URI**, so a refresh replacing the manifest under
the same URI left hashes computed for the previous snapshot describing the
new one. Keyed on the entry now, as the web screen already was.

**A failed "Load more" discarded the pages on screen and the retry cursor.**
Both are preserved now, and the success and failure paths share one
staleness-guarded `commit` helper so they cannot drift on which results they
may write — which is how they came to disagree.

Two nits: the six-skill counts. The docs sentence was more wrong than
reported (five awkward skills of six, three outright violations) and is
rewritten rather than renumbered.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ND46Ci64DC3tzBy9aG81Nc
Signed-off-by: cliffhall <cliff@futurescale.com>
This commit is contained in:
cliffhall 2026-09-07 18:43:58 -04:00
commit 775107776d
12 changed files with 452 additions and 56 deletions

View file

@ -78,6 +78,21 @@ describe("runMethod skills dispatch (#2248)", () => {
).rejects.toMatchObject({ exitCode: EXIT_CODES.USAGE });
});
it("rejects skills/get too when the server declares no extension", async () => {
// Declaring the extension commits a server to BOTH methods, so gating one
// and not the other is inconsistent with the thing being checked — and an
// undeclared server's -32601 is indistinguishable to a script from the
// -32602 a declared server returns for a URI it does not serve.
const client = mockClient({
getSkillsExtension: vi.fn().mockReturnValue(undefined),
getSkill: vi.fn(),
});
await expect(
runMethod(client, { method: "skills/get", uri: "skill://x/SKILL.md" }),
).rejects.toMatchObject({ exitCode: EXIT_CODES.USAGE });
expect(client.getSkill).not.toHaveBeenCalled();
});
it("keeps the { skill } envelope on skills/get", async () => {
// The client unwraps it for callers that want the entry; a CLI whose
// contract is "print the result" must not quietly reshape the wire form.

View file

@ -24,6 +24,28 @@ import type {
MethodOutcome,
} from "./method-types.js";
/**
* Refuse a `skills/*` call against a server that never declared the extension.
*
* Shared by `skills/list` and `skills/get` so the two cannot drift: declaring
* the extension commits a server to both, so a client that gates one and not
* the other is inconsistent with the thing it is checking. Not needed for
* `resources/directory/read`, whose stricter `directoryRead` gate lives in
* `InspectorClient` itself.
*/
function assertSkillsSupported(
inspectorClient: InspectorClient,
method: string,
): void {
if (!inspectorClient.getSkillsExtension()) {
throw new CliExitCodeError(
EXIT_CODES.USAGE,
`Server does not declare the ${SKILLS_EXTENSION_KEY} extension, so ${method} is not available.`,
{ code: "skills_unsupported" },
);
}
}
/**
* Run one MCP method against a connected {@link InspectorClient}.
* Core method dispatch used by the CLI (and other Inspector Node runners).
@ -299,13 +321,7 @@ export async function runMethod(
// list, which is right for a UI that must render *something*, and wrong
// for a CLI where "this server has no skills" and "this server does not
// serve skills at all" are different answers a script has to tell apart.
if (!inspectorClient.getSkillsExtension()) {
throw new CliExitCodeError(
EXIT_CODES.USAGE,
`Server does not declare the ${SKILLS_EXTENSION_KEY} extension, so ${args.method} is not available.`,
{ code: "skills_unsupported" },
);
}
assertSkillsSupported(inspectorClient, args.method);
managedSkillsState = new ManagedSkillsState(inspectorClient);
const skills = await managedSkillsState.refresh(args.metadata);
if (args.verify) {
@ -330,6 +346,12 @@ export async function runMethod(
"URI is required for skills/get method. Use --uri to specify the skill URI.",
);
}
// Same gate as `skills/list`, and for the same reason. Without it an
// undeclared server answers `-32601`, which a script cannot tell apart
// from the `-32602` a *declared* server returns for a skill URI it does
// not serve — "this server has no Skills support" and "no such skill"
// are different answers (Copilot).
assertSkillsSupported(inspectorClient, args.method);
const skill = await inspectorClient.getSkill(args.uri, args.metadata);
if (args.verify) {
const reports = await verifySkills(

View file

@ -806,6 +806,33 @@ describe("App (foundation)", () => {
await expectFrame(r, "Select a skill to view details");
});
it("leaves the Skills tab when the selected server does not serve it", async () => {
// The tab disappears from the bar when the gate goes false, but `activeTab`
// is independent of the bar — so without this the render branch keeps
// showing the pane for a server that never declared the extension, and the
// user is stranded on content they cannot navigate back to (Copilot).
h.ctrl.status = "connected";
h.ctrl.skillsExtension = { directoryRead: false };
const r = await mount(oneStdio());
await expectFrame(r, "Skills");
r.stdin.write("k");
await expectFrame(r, "Select a skill to view details");
// The server stops declaring it — the shape of switching to one without
// the extension, since the declaration is read off the live client.
h.ctrl.skillsExtension = undefined;
r.rerender(
<App
mcpServers={oneStdio()}
clientConfig={emptyClientConfig}
callbackUrlConfig={callbackUrlConfig}
/>,
);
await tick();
await expectFrame(r, "Server Configuration");
expect(r.lastFrame() ?? "").not.toContain("Select a skill to view details");
});
it("disconnects with 'd' when connected", async () => {
h.ctrl.status = "connected";
const { stdin } = await mount(oneStdio());

View file

@ -470,6 +470,100 @@ describe("SkillsTab (#2248)", () => {
await tick();
});
it("drops a verdict when the entry changes under the same URI", async () => {
// A refresh can replace the manifest or the frontmatter without the URI
// moving. A URI-keyed verdict would then present hashes and findings
// computed for the PREVIOUS snapshot as if they described the new one.
const digest = await sha256Digest(textToBytes(SKILL_MD));
const verifiable: SkillEntry = {
...clean,
resources: [
{
uri: "skill://clean/SKILL.md",
digest,
size: textToBytes(SKILL_MD).byteLength,
},
],
};
const { lastFrame, stdin, rerender } = render(
<SkillsTab
skills={[verifiable]}
pageCount={1}
inspectorClient={mockClient()}
width={140}
height={30}
focusedPane="list"
/>,
);
stdin.write(ENTER);
await tick();
expect(lastFrame() ?? "").toContain("Verified — Enter to re-verify");
// Same URI, different manifest — the old verdict must not carry over.
rerender(
<SkillsTab
skills={[
{
...verifiable,
resources: [
{ uri: "skill://clean/SKILL.md", digest: CLEAN_DIGEST, size: 9 },
],
},
]}
pageCount={1}
inspectorClient={mockClient()}
width={140}
height={30}
focusedPane="list"
/>,
);
await tick();
expect(lastFrame() ?? "").toContain(
"[Enter to verify digests and frontmatter]",
);
});
it("keeps a verdict across a reorder that leaves the entry unchanged", async () => {
// The reason the key is the entry rather than the list index: moving a
// skill down the list must not discard a verdict the user paid for.
const digest = await sha256Digest(textToBytes(SKILL_MD));
const verifiable: SkillEntry = {
...clean,
resources: [
{
uri: "skill://clean/SKILL.md",
digest,
size: textToBytes(SKILL_MD).byteLength,
},
],
};
const { lastFrame, stdin, rerender } = render(
<SkillsTab
skills={[verifiable]}
pageCount={1}
inspectorClient={mockClient()}
width={140}
height={30}
focusedPane="list"
/>,
);
stdin.write(ENTER);
await tick();
expect(lastFrame() ?? "").toContain("Verified — Enter to re-verify");
rerender(
<SkillsTab
skills={[verifiable, dynamic]}
pageCount={1}
inspectorClient={mockClient()}
width={140}
height={30}
focusedPane="list"
/>,
);
await tick();
expect(lastFrame() ?? "").toContain("Verified — Enter to re-verify");
});
it("falls back to the whole URI when a manifest entry has no path separator", async () => {
const odd: SkillEntry = {
uri: "skill://odd/SKILL.md",

View file

@ -626,6 +626,28 @@ function App({
!!selectedInspectorClient?.getSkillsExtension() &&
inspectorStatus === "connected";
// Switch away from the Skills tab when the selected server does not serve it.
//
// The same handling the Auth tab gets above, and needed for the same reason:
// the tab disappears from the bar when the gate goes false, but `activeTab`
// is independent of the bar, so the render branch would keep showing the pane
// for a server that never declared the extension — content the user can see
// but can no longer navigate back to (Copilot).
//
// Gated on `connected` rather than on the extension alone: the declaration is
// only knowable after the handshake, so resetting while a reconnect is in
// flight would bounce the user off the tab they were reading and not return
// them to it.
useEffect(() => {
if (
activeTab === "skills" &&
inspectorStatus === "connected" &&
!showSkillsTab
) {
setActiveTab("info");
}
}, [activeTab, inspectorStatus, showSkillsTab]);
// Connect — on 401 or mid-session auth recovery, run OAuth then retry.
type TuiOAuthRunResult =
| "success"

View file

@ -64,6 +64,18 @@ const FILE_COLOR: Record<string, string> = {
"read-error": "red",
};
/**
* What a verification result is a result *about*: the whole entry, serialized.
*
* `JSON.stringify` is enough here — this compares an entry against a later copy
* of *itself* from the same server, so key order is stable and there is no need
* for the canonical form `skillEntriesMatch` uses to compare two independently
* produced entries.
*/
function entryKey(entry: SkillEntry): string {
return JSON.stringify(entry);
}
/** The file name a manifest URI ends in, for a list that must fit 40 columns. */
function fileNameOf(uri: string): string {
const cut = uri.lastIndexOf("/");
@ -100,14 +112,23 @@ export function SkillsTab({
const [error, setError] = useState<string | null>(null);
const [verifying, setVerifying] = useState(false);
/**
* The last verification, keyed by the skill URI it was run for. Keyed rather
* than cleared on selection change so moving off a skill and back does not
* silently discard a verdict the user just paid a round trip for — and keyed
* by URI rather than index so a refresh that reorders the list cannot show
* one skill's verdict under another's name.
* The last verification, keyed by the **entry it was computed against**.
*
* Keyed rather than cleared on selection change, so moving off a skill and
* back does not silently discard a verdict the user just paid a round trip
* for. Keyed by a serialization of the entry rather than by its index, so a
* refresh that reorders the list cannot show one skill's verdict under
* another's name — and rather than by its URI alone, because a refresh can
* replace the manifest or the frontmatter *under the same URI*, and a
* URI-keyed verdict would then present hashes and findings computed for the
* previous snapshot as if they described the new one (Copilot).
*
* The same key the web screen uses, for the same reason: re-verifying after a
* metadata-only refresh is the cheap direction to be wrong in; showing a
* verdict computed against a different entry is not.
*/
const [report, setReport] = useState<{
uri: string;
key: string;
result: SkillVerifyReport;
} | null>(null);
const scrollViewRef = useRef<ScrollViewRef>(null);
@ -124,7 +145,7 @@ export function SkillsTab({
void (async () => {
try {
const [result] = await verifySkills(inspectorClient, [skill]);
setReport({ uri: skill.uri, result });
setReport({ key: entryKey(skill), result });
} catch (err) {
if (err instanceof AuthRecoveryRequiredError) {
onAuthRecoveryRequired?.(err);
@ -196,7 +217,9 @@ export function SkillsTab({
const detailWidth = width - listWidth;
const issues = selectedSkill ? checkSkillConformance(selectedSkill) : [];
const activeReport =
selectedSkill && report?.uri === selectedSkill.uri ? report.result : null;
selectedSkill && report?.key === entryKey(selectedSkill)
? report.result
: null;
const manifest =
selectedSkill && selectedSkill.resources !== DYNAMIC_RESOURCES
? selectedSkill.resources

View file

@ -1784,6 +1784,27 @@ describe("SkillsScreen directory browsing (#2248)", () => {
});
}
it("starts collapsed, since its content needs a round trip nobody has made", async () => {
// Open, it would hold a button and an empty frame — advertising content
// that is not there while taking height from the sections that have some.
const user = userEvent.setup();
renderWithMantine(
<ControlledSkillsScreen
onReadResourceDirectory={directoryReader({ [ROOT]: { resources: [] } })}
/>,
);
await user.click(screen.getByText("data-analysis"));
expect(screen.getByRole("button", { name: /Directory/ })).toHaveAttribute(
"aria-expanded",
"false",
);
// Still reachable, and the other sections are unaffected.
expect(screen.getByRole("button", { name: /Resources/ })).toHaveAttribute(
"aria-expanded",
"true",
);
});
it("renders no Directory section when the server did not declare directoryRead", async () => {
const user = userEvent.setup();
renderWithMantine(<ControlledSkillsScreen />);
@ -1808,6 +1829,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
await user.click(screen.getByText("data-analysis"));
expect(onReadResourceDirectory).not.toHaveBeenCalled();
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByTestId("skill-directory")).toBeInTheDocument(),
@ -1836,6 +1858,8 @@ describe("SkillsScreen directory browsing (#2248)", () => {
<ControlledSkillsScreen onReadResourceDirectory={reader} />,
);
await user.click(screen.getByText("data-analysis"));
// Directory starts collapsed — see `DEFAULT_OPEN_SECTIONS`.
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByTestId("skill-directory")).toBeInTheDocument(),
@ -1906,6 +1930,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
/>,
);
await user.click(screen.getByText("data-analysis"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByTestId("skill-directory")).toBeInTheDocument(),
@ -1934,6 +1959,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
/>,
);
await user.click(screen.getByText("data-analysis"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(
@ -2024,6 +2050,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
<ControlledSkillsScreen onReadResourceDirectory={reader} />,
);
await user.click(screen.getByText("dynamic-report"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByTestId("skill-directory")).toBeInTheDocument(),
@ -2044,6 +2071,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
/>,
);
await user.click(screen.getByText("data-analysis"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByText("This directory is empty.")).toBeInTheDocument(),
@ -2063,6 +2091,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
/>,
);
await user.click(screen.getByText("data-analysis"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await waitFor(() =>
expect(screen.getByText(/Not a directory resource/)).toBeInTheDocument(),
@ -2072,6 +2101,41 @@ describe("SkillsScreen directory browsing (#2248)", () => {
).toBeInTheDocument();
});
it("keeps the pages already shown when Load more fails, and can retry", async () => {
// Replacing the state outright made the table vanish and stranded the
// reader with no way back to that page short of restarting at the root.
const user = userEvent.setup();
let fail = true;
const onReadResourceDirectory = vi.fn(
async (_uri: string, cursor?: string) => {
if (cursor === undefined) {
return { resources: [CHILD_FILE], nextCursor: "1" } as never;
}
if (fail) {
fail = false;
throw new Error("page two exploded");
}
return { resources: [CHILD_DIR] } as never;
},
);
await openRoot(user, onReadResourceDirectory as never);
await user.click(screen.getByRole("button", { name: "Load more" }));
await waitFor(() =>
expect(screen.getByText(/page two exploded/)).toBeInTheDocument(),
);
// The first page is still on screen…
expect(
within(screen.getByTestId("skill-directory")).getByText(CHILD_FILE.uri),
).toBeInTheDocument();
// …and the cursor survived, so the same page can be retried.
await user.click(screen.getByRole("button", { name: "Load more" }));
await waitFor(() => {
const table = within(screen.getByTestId("skill-directory"));
expect(table.getByText(CHILD_FILE.uri)).toBeInTheDocument();
expect(table.getByText(CHILD_DIR.uri)).toBeInTheDocument();
});
});
it("drops a listing when the selection changes mid-read", async () => {
// A read still in flight when the user switches skills must not land
// afterwards and paint one skill's tree under another's name.
@ -2089,6 +2153,7 @@ describe("SkillsScreen directory browsing (#2248)", () => {
/>,
);
await user.click(screen.getByText("data-analysis"));
await user.click(screen.getByRole("button", { name: /Directory/ }));
await user.click(screen.getByRole("button", { name: "Read directory" }));
await user.click(screen.getByText("right-name"));
release?.({ resources: [CHILD_FILE] });

View file

@ -495,6 +495,27 @@ const ALL_SECTIONS = [
"resource",
];
/**
* The sections that open by default — everything except Directory.
*
* Directory is the one section whose content requires a round trip the user has
* not made yet, so open it holds a button and an empty frame: it advertises
* content that is not there, while taking height from the sections that do have
* some. With five open sections in a short pane each is squeezed to its floor
* and scrolls internally, which is the documented fallback but a poor first
* impression — and the one it costs most is the file viewer, the section
* `viewerFlex` exists to give the remainder to.
*
* The same argument as Conformance's auto-collapse, one step earlier: that one
* closes a section whose header already carries the whole answer, this one
* closes a section that has no answer yet. Both are defaults; neither prevents
* opening it, and `openSections` outlives a selection, so a user who opens
* Directory keeps it open across skills.
*/
const DEFAULT_OPEN_SECTIONS = ALL_SECTIONS.filter(
(section) => section !== "directory",
);
/**
* The open set for the FIRST render.
*
@ -508,13 +529,13 @@ function initialOpenSections(
skills: SkillEntry[],
selectedSkillUri: string | undefined,
): string[] {
if (selectedSkillUri === undefined) return ALL_SECTIONS;
if (selectedSkillUri === undefined) return DEFAULT_OPEN_SECTIONS;
const wanted = skillUriIdentity(selectedSkillUri);
const entry = skills.find((skill) => skillUriIdentity(skill.uri) === wanted);
if (entry === undefined) return ALL_SECTIONS;
if (entry === undefined) return DEFAULT_OPEN_SECTIONS;
return checkSkillConformance(entry).length > 0
? ALL_SECTIONS
: ALL_SECTIONS.filter((section) => section !== "conformance");
? DEFAULT_OPEN_SECTIONS
: DEFAULT_OPEN_SECTIONS.filter((section) => section !== "conformance");
}
/**
@ -1032,11 +1053,17 @@ export function SkillsScreen({
(uri: string, key: string, cursor?: string) => {
if (!onReadResourceDirectory) return;
const attempt = (nextAttempt.current += 1);
const write = (next: Omit<DirectoryState, "key" | "attempt">) =>
/**
* Commit a settled result, dropping it when it no longer belongs to the
* pane on screen — a different skill, or a newer read of this one.
*/
const commit = (
next: (prev: DirectoryState) => Omit<DirectoryState, "key" | "attempt">,
) =>
setDirectory((prev) => {
if (prev.key !== null && prev.key !== key) return prev;
if (prev.attempt !== undefined && prev.attempt > attempt) return prev;
return { key, attempt, ...next };
return { key, attempt, ...next(prev) };
});
// The path is claimed before the request goes out, so the header names
// the directory being read rather than continuing to announce the
@ -1058,26 +1085,28 @@ export function SkillsScreen({
// `catch`, which surfaces the message in the section.
void onReadResourceDirectory(uri, cursor)
.then((page) => {
setDirectory((prev) => {
if (prev.key !== null && prev.key !== key) return prev;
if (prev.attempt !== undefined && prev.attempt > attempt) {
return prev;
}
const held = cursor === undefined ? [] : (prev.children ?? []);
return {
key,
attempt,
uri,
children: [...held, ...page.resources],
nextCursor: page.nextCursor,
};
});
commit((prev) => ({
uri,
children: [
...(cursor === undefined ? [] : (prev.children ?? [])),
...page.resources,
],
nextCursor: page.nextCursor,
}));
})
.catch((err: unknown) => {
write({
// A FAILED page leaves what is already on screen where it is, and
// keeps the cursor that would retry it. Replacing the state outright
// made the table vanish and stranded the reader with no way back to
// that page short of restarting at the root (Copilot). Only a first
// read of a directory has nothing to preserve.
commit((prev) => ({
uri,
...(cursor === undefined
? {}
: { children: prev.children, nextCursor: cursor }),
message: err instanceof Error ? err.message : String(err),
});
}));
});
},
[onReadResourceDirectory],

View file

@ -137,13 +137,87 @@ describe("verifySkills (#2248)", () => {
expect(report.ok).toBe(false);
});
it("refuses a block for a DIFFERENT uri rather than verifying it", async () => {
// The dangerous shape, and the reason positional selection is wrong: these
// bytes are about to be hashed against THIS file's advertised digest, so
// accepting a block the server labelled something else verifies one file's
// content against another file's digest — and can report that as
// `verified`. A false pass is worse than a missing check.
const skill = await entry();
const readResource = vi.fn(async () => ({
result: {
contents: [{ uri: "skill://demo/unrelated.md", text: "other bytes" }],
},
}));
const client = { readResource } as unknown as InspectorClientProtocol;
const [report] = await verifySkills(client, [skill]);
expect(report.files.every((f) => f.status === "read-error")).toBe(true);
expect(report.files[0].reason).toMatch(/no content block for this URI/);
expect(report.ok).toBe(false);
});
it("finds the matching block when it is not the first one", async () => {
// A server may answer with more than one block, in any order; taking
// `contents[0]` would hash the wrong file's bytes.
const bytes = new TextEncoder().encode(REF);
const skill: SkillEntry = {
uri: "skill://demo/SKILL.md",
frontmatter: { name: "demo", description: "A demo" },
resources: [
{
uri: "skill://demo/ref.md",
digest: await sha256Digest(bytes),
size: bytes.byteLength,
},
],
};
const readResource = vi.fn(async () => ({
result: {
contents: [
{ uri: "skill://demo/decoy.md", text: "decoy" },
{ uri: "skill://demo/ref.md", text: REF },
],
},
}));
const client = { readResource } as unknown as InspectorClientProtocol;
const [report] = await verifySkills(client, [skill]);
expect(report.files[0].status).toBe("verified");
});
it("ignores a malformed block while still finding the real one", async () => {
const bytes = new TextEncoder().encode(REF);
const skill: SkillEntry = {
uri: "skill://demo/SKILL.md",
frontmatter: { name: "demo", description: "A demo" },
resources: [
{
uri: "skill://demo/ref.md",
digest: await sha256Digest(bytes),
size: bytes.byteLength,
},
],
};
const readResource = vi.fn(async () => ({
result: {
contents: [
null,
{ uri: 42 },
{ uri: "skill://demo/ref.md", text: REF },
],
},
}));
const client = { readResource } as unknown as InspectorClientProtocol;
const [report] = await verifySkills(client, [skill]);
expect(report.files[0].status).toBe("verified");
});
it("reports a response with no content blocks as a read failure", async () => {
const skill = await entry();
const readResource = vi.fn(async () => ({ result: { contents: [] } }));
const client = { readResource } as unknown as InspectorClientProtocol;
const [report] = await verifySkills(client, [skill]);
expect(report.files[0]).toMatchObject({ status: "read-error" });
expect(report.files[0].reason).toMatch(/no content blocks/);
expect(report.files[0].reason).toMatch(/no content block for this URI/);
});
it("reports a block carrying neither text nor blob as a read failure", async () => {

View file

@ -121,8 +121,8 @@ describe("Skills extension over a real transport (#2234)", () => {
// below — so a modern page missing the envelope surfaces here as a
// rejection rather than as a missing property.
//
// The fixture pages at two over five skills, so a client that stops
// here sees less than half.
// The fixture pages at two over six skills, so a client that stops
// here sees a third of the catalog.
expect(first.skills).toHaveLength(2);
expect(first.nextCursor).toBeDefined();

View file

@ -32,6 +32,7 @@ import {
checkSkillFrontmatterMatch,
skillDisplayName,
skillFileBytes,
skillUriIdentity,
verifySkillResource,
type SkillIssue,
type SkillVerification,
@ -85,19 +86,38 @@ interface ReadContents {
}
/**
* The first content block of a `resources/read` result.
* The block of a `resources/read` result that answers for `uri` — **selected by
* URI, never by position**.
*
* `contents[0]` rather than a search by URI: a server may legitimately answer
* with a canonicalized spelling of the URI we asked for, and matching on the
* string would reject it. A result with no blocks is a read failure and is
* reported as one.
* `contents` is an array, and taking `contents[0]` is wrong in the one way that
* matters here: these bytes are about to be hashed against `uri`'s advertised
* digest, so accepting a block the server labelled something else would verify
* one file's content against another file's digest — and could report that as
* `verified`. A false pass from a positional read is worse than a missing
* check, because it is an affirmative statement about a file nobody looked at.
*
* A **normalized** match is accepted, because a server may echo the URI back in
* a different but equivalent form — a resolved `..`, a percent-encoding
* difference. That is what `skillUriIdentity` is for, and it is the same rule
* the whole module applies to every other URI comparison, so a server cannot be
* treated as conforming by one check and non-conforming by another.
*
* `undefined` when nothing answers for the URI, which the caller reports as a
* read failure. This mirrors `onReadSkillFile` in the web client, deliberately:
* two code paths that hash bytes against a digest must not disagree about which
* bytes they are.
*/
function firstContents(result: unknown): ReadContents | undefined {
function contentsFor(result: unknown, uri: string): ReadContents | undefined {
const contents = (result as { contents?: unknown })?.contents;
if (!Array.isArray(contents) || contents.length === 0) return undefined;
const first: unknown = contents[0];
if (typeof first !== "object" || first === null) return undefined;
return first as ReadContents;
if (!Array.isArray(contents)) return undefined;
const wanted = skillUriIdentity(uri);
for (const block of contents) {
if (typeof block !== "object" || block === null) continue;
const got = (block as { uri?: unknown }).uri;
if (typeof got !== "string") continue;
if (skillUriIdentity(got) === wanted) return block as ReadContents;
}
return undefined;
}
/**
@ -136,7 +156,7 @@ export async function verifySkills(
let contents: ReadContents | undefined;
try {
const invocation = await client.readResource(resource.uri, metadata);
contents = firstContents(invocation.result);
contents = contentsFor(invocation.result, resource.uri);
} catch (err) {
if (err instanceof AuthRecoveryRequiredError) throw err;
files.push({
@ -150,7 +170,8 @@ export async function verifySkills(
files.push({
uri: resource.uri,
status: "read-error",
reason: "resources/read returned no content blocks.",
reason:
"resources/read returned no content block for this URI, so there are no bytes that can be checked against its digest.",
});
continue;
}
@ -179,7 +200,7 @@ export async function verifySkills(
if (entryText === undefined) {
try {
const invocation = await client.readResource(entry.uri, metadata);
const contents = firstContents(invocation.result);
const contents = contentsFor(invocation.result, entry.uri);
if (typeof contents?.text === "string") entryText = contents.text;
} catch (err) {
// Left undefined: the frontmatter check is skipped below. When the

View file

@ -88,10 +88,14 @@ It works on **either era**: `skills/list`, `skills/get` and
era codec defines, so the SDK's era gate skips them entirely — which is why
this fixture, unlike the tasks ones, needs no per-era variant.
Four of the six skills are deliberately awkward, because the checks the Skills
tab runs are untestable without them. Only two are actual violations — the
Five of the six skills are deliberately awkward, because the checks the Skills
tab runs are untestable without them. Only **three** are outright violations
(`tampered-notes`, `lying-listing`, `wrong-folder` — the three `--verify` fails
on). The other two are subtler and neither is an error on its own: the
`"dynamic"` form is **conforming**, and is here because "legal but unverifiable"
is the case most easily buried:
is the case most easily buried; and `stale-manifest`'s entry is fully conforming
too, with the defect living in the disagreement between its manifest and its
directory listing:
| Skill | What it exercises |
| --- | --- |