Settings → MCP panel #172

Merged
lz merged 7 commits from feat/oauth-settings into main 2026-09-17 13:58:52 +02:00
Owner

Closes #153. Stacked on #170 — base is feat/mcp-resource-server, not main. Most of what this panel reports did not exist before #170.

Design record: docs/superpowers/specs/2026-09-17-mcp-settings-panel-design.md. AGENTS.md gains fact #39.

The problem

Everything MCP needs shipped and none of it was reachable. /api/settings/mcp/clients has served the CIMD allowlist since #152 with no UI over it, so allowing a client meant a SQL insert. The connector URL an operator pastes into Claude was derived and never displayed. And the reasons MCP can refuse to serve were computed, phrased for operators, and visible only as the body of a 503 nobody reads.

The milestone's goal is driving Nexus from a phone. An operator on a phone cannot run SQL.

What this builds

Two cards. The server card shows the connector URL — taken from the derived string, never recomposed from an origin and MCP_PATH, because the token is audience-bound to it — plus three checks and an action link where there is something to do. The clients card is the allowlist, and says plainly that this is who may register, not who is connected, and that Nexus never fetches these URLs.

The three checks are rendered in every state, including a healthy one. A single reason string was the first design and it was wrong: it only ever speaks when MCP is broken, so an operator looking at a working instance learns nothing about why it works or which lever changes that. The panel would teach the system exclusively by failing.

serving is a function of publicUrl and signingKey only. An empty allowlist is warn, never error: verification is JWT-based and never consults the allowlist, so an existing token keeps working — what an empty allowlist stops is a new client registering. Folding it into serving reports a working endpoint as down, and is the plausible, cautious-looking, wrong change.

Three of #153's four sections are not built — see #171

Each describes a seam that does not exist; building them would ship controls that lie.

  • Per-entry reachability chips — refused permanently. Resolving an operator-supplied URL server-side puts an SSRF primitive next to the Docker API on 127.0.0.1:2375 (Nexus runs network_mode: host). The allowlist being an exact string comparison is the security property. Chips on our own .well-known routes were considered and dropped separately — safe, but decided by the same two facts the checks already report, and fetched from an origin that need not be the one Claude uses.
  • Authorization-server selector — allowDcr is hardcoded false, there is no issuer setting, and deriveIssuer always returns Nexus's own origin. A segmented control with one reachable position.
  • Connected clients / revoke — cimd_clients is (url, added_at). That data lives in grants.

Two corrections to #153's body

  • Token TTL is 5 minutes, not 1h (accessTokenTTL: 300).
  • Removal revokes nothing. It stops the next authorization. The confirmation says so, and says "until 14 days after it first authorized" — an absolute deadline from the original grant, because each rotation inherits its predecessor's remainingTTL. "For up to 14 days" reads as 14 days from now and is wrong for every client except a brand-new one.

Also fixed

setOauthJwks now carries why the keyset is null. Fact #38 keeps "unset" and "broken" apart at boot — the loader returns null for one and throws for the other — but both arrived at the singleton as a bare null, leaving the distinction only in the log. Reported to an operator that becomes "not set" for a variable they have set and mistyped.

public_url's help text now says MCP derives its connector URL and issuer from it — that setting is where this panel's Change link lands, and it is the most common reason MCP is not serving.

Verification

Gates run separately, never chained: typecheck 0 errors · 2190 tests · lint clean · pnpm build 0.

Guards proven to fail, each mutation diffed against a backup copy first to confirm it landed:

  • folding allowedClients into serving → KILLED both named tests, and only those two (17 others still passed, so it was targeted rather than a blanket break)
  • removing a field from either client DTO → KILLED the type-level parity test at the exact line

Driven in a browser against a real instance (node build/server.js, seeded DB, minted signing key) — serving, no-public-URL and no-signing-key states, add, client-side refusal of an invalid URL, remove with confirmation, both 207 paths forced via route mocking, and 390px.

That pass found four things the suite structurally cannot (vitest runs node with no DOM):

  • word-break: break-all split prose mid-word (Settings → Instanc/e)
  • a seventh tab made the strip wrap inside its buttons rather than between them
  • the remove dialog's Escape handler was dead — it sits on the dialog, but opening it left focus on the row's Remove button, outside it, so the keydown bubbled up a different branch. Measured: document.activeElement was BUTTON.btn sm
  • my own "no request was sent" check was worthless — the server logs no requests at all, so the grep proved nothing; re-verified through the browser's network log

Review

/simplify ran four agents. Efficiency and simplification: clean. Three fixes applied — one decision point for the signing key (checkSigningKey, removing a duplicated error literal between the 503 and the panel), parity tests for both wire-duplicated DTOs rather than one, and reuse of app.css's shared .overlay instead of a third hand-rolled backdrop.

Two choices were reviewed and confirmed correct at the depth made: the setOauthJwks(jwks, fault?) shape, and declaring McpStatusDTO on both sides of the wire.

Follow-up, not in this PR: PassphraseRotationModal.svelte has the same dead-Escape bug, and a shared Modal.svelte would fix it in every copy at once. Being done separately.

Closes #153. Stacked on #170 — **base is `feat/mcp-resource-server`, not `main`.** Most of what this panel reports did not exist before #170. Design record: `docs/superpowers/specs/2026-09-17-mcp-settings-panel-design.md`. AGENTS.md gains fact #39. ## The problem Everything MCP needs shipped and none of it was reachable. `/api/settings/mcp/clients` has served the CIMD allowlist since #152 with no UI over it, so allowing a client meant a SQL insert. The connector URL an operator pastes into Claude was derived and never displayed. And the reasons MCP can refuse to serve were computed, phrased for operators, and visible only as the body of a 503 nobody reads. The milestone's goal is driving Nexus from a phone. An operator on a phone cannot run SQL. ## What this builds **Two cards.** The server card shows the connector URL — taken from the derived string, never recomposed from an origin and `MCP_PATH`, because the token is audience-bound to it — plus three checks and an action link where there is something to do. The clients card is the allowlist, and says plainly that this is who *may* register, not who is connected, and that Nexus never fetches these URLs. **The three checks are rendered in every state, including a healthy one.** A single `reason` string was the first design and it was wrong: it only ever speaks when MCP is broken, so an operator looking at a working instance learns nothing about why it works or which lever changes that. The panel would teach the system exclusively by failing. **`serving` is a function of `publicUrl` and `signingKey` only.** An empty allowlist is `warn`, never `error`: verification is JWT-based and never consults the allowlist, so an existing token keeps working — what an empty allowlist stops is a *new* client registering. Folding it into `serving` reports a working endpoint as down, and is the plausible, cautious-looking, wrong change. ## Three of #153's four sections are not built — see #171 Each describes a seam that does not exist; building them would ship controls that lie. - **Per-entry reachability chips — refused permanently.** Resolving an operator-supplied URL server-side puts an SSRF primitive next to the Docker API on `127.0.0.1:2375` (Nexus runs `network_mode: host`). The allowlist being an exact string comparison *is* the security property. Chips on our own `.well-known` routes were considered and dropped separately — safe, but decided by the same two facts the checks already report, and fetched from an origin that need not be the one Claude uses. - **Authorization-server selector** — `allowDcr` is hardcoded `false`, there is no issuer setting, and `deriveIssuer` always returns Nexus's own origin. A segmented control with one reachable position. - **Connected clients / revoke** — `cimd_clients` is `(url, added_at)`. That data lives in grants. ## Two corrections to #153's body - Token TTL is **5 minutes**, not 1h (`accessTokenTTL: 300`). - **Removal revokes nothing.** It stops the next authorization. The confirmation says so, and says *"until 14 days after it first authorized"* — an absolute deadline from the original grant, because each rotation inherits its predecessor's `remainingTTL`. "For up to 14 days" reads as 14 days from *now* and is wrong for every client except a brand-new one. ## Also fixed `setOauthJwks` now carries **why** the keyset is null. Fact #38 keeps "unset" and "broken" apart at boot — the loader returns null for one and throws for the other — but both arrived at the singleton as a bare null, leaving the distinction only in the log. Reported to an operator that becomes "not set" for a variable they *have* set and mistyped. `public_url`'s help text now says MCP derives its connector URL and issuer from it — that setting is where this panel's `Change` link lands, and it is the most common reason MCP is not serving. ## Verification Gates run **separately**, never chained: typecheck 0 errors · **2190 tests** · lint clean · `pnpm build` 0. **Guards proven to fail**, each mutation diffed against a backup copy first to confirm it landed: - folding `allowedClients` into `serving` → KILLED both named tests, and only those two (17 others still passed, so it was targeted rather than a blanket break) - removing a field from either client DTO → KILLED the type-level parity test at the exact line **Driven in a browser** against a real instance (`node build/server.js`, seeded DB, minted signing key) — serving, no-public-URL and no-signing-key states, add, client-side refusal of an invalid URL, remove with confirmation, **both 207 paths** forced via route mocking, and 390px. That pass found four things the suite structurally cannot (vitest runs node with no DOM): - `word-break: break-all` split prose mid-word (`Settings → Instanc/e`) - a seventh tab made the strip wrap *inside* its buttons rather than between them - the remove dialog's **Escape handler was dead** — it sits on the dialog, but opening it left focus on the row's Remove button, outside it, so the keydown bubbled up a different branch. Measured: `document.activeElement` was `BUTTON.btn sm` - my own "no request was sent" check was worthless — the server logs no requests at all, so the grep proved nothing; re-verified through the browser's network log ## Review `/simplify` ran four agents. Efficiency and simplification: clean. Three fixes applied — one decision point for the signing key (`checkSigningKey`, removing a duplicated error literal between the 503 and the panel), parity tests for **both** wire-duplicated DTOs rather than one, and reuse of `app.css`'s shared `.overlay` instead of a third hand-rolled backdrop. Two choices were reviewed and confirmed correct at the depth made: the `setOauthJwks(jwks, fault?)` shape, and declaring `McpStatusDTO` on both sides of the wire. **Follow-up, not in this PR:** `PassphraseRotationModal.svelte` has the same dead-Escape bug, and a shared `Modal.svelte` would fix it in every copy at once. Being done separately.
lz added 7 commits 2026-09-17 13:30:02 +02:00
The Settings → MCP panel needs to refuse a bad metadata-document URL before
the round trip, and to show InvalidAllowlistUrl's own prose rather than
paraphrasing it -- but a Svelte component cannot import $server/cimd/url.

Moved to src/lib/cimd-url.ts, the documented precedent being git-ref.ts,
which sits outside $lib/server for exactly this reason (the palette and the
sidebar refuse an invalid branch name client-side).

This is a relocation, not a second declaration: cimd/service.ts still calls
the same function on the way to the insert, so the server stays
authoritative and client-side validation is a latency nicety, never a safety
boundary. The both-sides-of-the-wire parity rule therefore does not apply --
the parity test moves with the file and keeps doing its existing job of
pinning the mirror against oidc-provider's isValidClientIdUrl.
GET /api/settings/mcp/status, vault-locked like its sibling. Every reason it
returns is produced by the module that owns the fault -- mcp/metadata.ts,
jwks.ts, or the boot loader -- so the panel and the 503 cannot describe the
same failure differently.

Three checks, reported in EVERY state including a healthy one. A single
`reason` string was the first design and it was wrong: it only ever speaks
when MCP is broken, so an operator looking at a working instance learns
nothing about why it works or which lever changes that. The panel would
teach the system exclusively by failing.

`serving` is a function of publicUrl and signingKey ONLY. An empty allowlist
is `warn`, never `error`: verification is JWT-based and never consults the
allowlist, so an existing token keeps working -- what an empty allowlist
stops is a NEW client registering. Folding it into `serving` reports a
working endpoint as down, and is the plausible, cautious-looking, wrong
change. status-parity.test.ts drives mcpStatus and mcpRuntime over one input
matrix and demands they agree; proven to fail by making exactly that change.

setOauthJwks now carries WHY the keyset is null. Fact #38 keeps "unset" and
"broken" apart at boot -- the loader returns null for one and throws for the
other -- but both arrived at the singleton as a bare null, leaving the
distinction only in the log. Reported to an operator that becomes "not set"
for a variable they HAVE set and mistyped, sending them to fix the one thing
already done. The two are set through one function so a fault cannot outlive
the null it explains.
Everything MCP needs has shipped and none of it was reachable: the allowlist
API has had no UI since #152, so allowing a client meant a SQL insert, and
the connector URL an operator must paste into Claude was derived and never
displayed. The milestone's goal is driving Nexus from a phone, and an
operator on a phone cannot run SQL.

Two cards. The server card shows the connector URL -- taken from the derived
string, never recomposed from an origin and MCP_PATH, because the token is
audience-bound to it -- and the three checks with an action link where there
is something to do. The clients card is the allowlist, and says plainly that
this is who MAY register, not who is connected, and that Nexus never fetches
these URLs.

Both mutations can return 207, and call() RESOLVES it (res.ok covers
200-299), so the panel branches on the body. Both operations SUCCEEDED: a
POST 207 keeps the row and does not invite a retry, a DELETE 207 says the
client may still authenticate rather than reporting a completed revocation.
Rendered as a warning note, never as an error. removeCimdClient is typed as
the union rather than Promise<void>, which compiles and silently discards
the warning.

The remove confirmation says removal does NOT disconnect, "until 14 days
after it first authorized" -- an absolute deadline from the original grant,
because each rotation inherits its predecessor's remainingTTL. "For up to 14
days" reads as 14 days from now and is wrong for every client except a
brand-new one.

Two layout bugs found by driving a browser, neither visible to the suite:
word-break: break-all split prose mid-word ("Settings → Instanc/e"), and a
seventh tab made the strip wrap inside its buttons rather than between them.

public_url's help text now says MCP derives its connector URL and issuer
from it -- that setting is where this panel's Change link lands, and it is
the most common reason MCP is not serving.
#153 named four sections; two had something behind them. The design doc says
why the other two are not built rather than faked, and #171 tracks them.

The refusal worth having in AGENTS.md is the reachability chip: resolving an
operator-supplied URL server-side puts an SSRF primitive beside the Docker
API on 127.0.0.1:2375, and the allowlist being an exact string comparison IS
the security property. Chips on our own .well-known routes were considered
and dropped separately -- safe, but decided by the same two facts the checks
already report, and fetched from an origin that need not be the one Claude
uses.
mcpRuntime and mcpStatus each spelled out the same publicJwkFor try/catch and
the same "the OAuth signing keyset is unusable" literal, so the 503 and the
Settings panel could drift into describing one fault two ways. checkSigningKey
in oauth/jwks.ts -- which already owns publicJwkFor and NO_SIGNING_KEY -- now
owns the verdict and its wording, and both read it.

Only the verdict is shared. The control flows stay separate because they
genuinely differ: mcpRuntime short-circuits on the first failure for a single
caller, while mcpStatus is deliberately eager and reports every check even on
a healthy instance.

`fault` stays a parameter rather than being read inside, because only the
status caller can tell "unset" from "broken" (fact #38); a caller without that
information gets the generic NO_SIGNING_KEY, which is the honest answer for it.

Found by the simplification pass.
This change duplicated two types across the $lib/server boundary --
McpStatusDTO and CimdClientDTO -- and guarded only the first. A guard covering
one of two identical risks is worse than no guard: it reads as though the
pattern is handled.

Both now live in one file, in lib/api/ beside worker-dto-parity.test.ts, which
is the established home and precedent for this (WorkerDTO is duplicated and
pinned the same way).

Proven to fail: removing `added_at` from the client CimdClientDTO makes
typecheck name mcp-dto-parity.test.ts:41. Both assignment directions are kept,
because one alone permits the other side to be MISSING a field -- which is
exactly the drift that leaves a column blank with a clean build.

Also fixes a doubled prefix in cimd-url.ts's header: two overlapping sed
patterns during the move rewrote `url-parity.test.ts` into
`cimd-cimd-url-parity.test.ts`.
fix(settings): reuse the shared overlay, and make the dialog's Escape work
Some checks failed
ci / nexus (pull_request) Failing after 13m26s
ci / images (pull_request) Has been skipped
pr-image-cleanup / delete-pr-images (pull_request) Successful in 18s
eae9a1d2b0
Two findings against the remove-confirmation dialog.

app.css already provides `.overlay`/`.overlay-box` -- listed in its own SCOPE
header as shared vocabulary, used by LogsModal and InstallSkillModal. This
dialog hand-rolled a `.modal-backdrop`/`.modal` copy instead, which would have
made three independent backdrops in components/ (NewWorkerModal and
PassphraseRotationModal are the other two). Reusing the shared classes deletes
that copy and takes the count back to two without touching shipped components.
`.btn-row` comes from app.css too; only the dialog's max-width is its own.

The Escape handler was DEAD CODE. It sits on the dialog element, but opening
the dialog leaves focus on the row's Remove button -- outside it -- so the
keydown bubbles up a different branch and the handler never fires. Measured
before the fix: document.activeElement was `BUTTON.btn sm` and Escape did
nothing. The dialog now takes focus when it opens, which is also the right
thing to do for a modal irrespective of key handling.

Verified in a browser both ways: focus is now `DIV.overlay show`, Escape
closes, and a real backdrop click still closes. Neither is reachable by the
test suite -- vitest runs node with no DOM.

The same dead-handler pattern exists in PassphraseRotationModal.svelte and is
left alone here as outside this change.
Author
Owner

Review from the session that built #170, which this is stacked on. I checked the things I am uniquely placed to check — whether #170's guarantees survive your changes — rather than re-reviewing what your own four agents already covered.

#170's guards all still bite under this branch

This PR modifies code #170 owns (deps.ts, jwks.ts, singletons.ts, server.ts), including removing a duplicated error literal between the 503 and the panel. That touches the anchors of the mutation gate #170 shipped on, so I re-ran it against eae9a1d:

baseline: 8 target suites green before mutating

KILLED  audience         KILLED  origin-leg
KILLED  issuer           KILLED  https-check
KILLED  typ              KILLED  key-exclusive
KILLED  scope-admit      KILLED  key-single
KILLED  tool-filter      KILLED  key-curve
KILLED  host-order       KILLED  gate-placement

12/12 guards bite.

Audience/issuer/typ binding, the tool-surface admission rule, the registry filter, Host-before-bearer ordering, the Origin leg, the HTTPS refusal, both-spellings exclusivity, single-key and EC P-256, and the gate placement that keeps a vault key off /api/mcp — all still kill their named tests. Full suite on your head: 195 files / 2190 tests, exit 0, matching your claim exactly.

A false alarm I caused, in case you saw anything odd: my first run reported 0/12 — all twelve SURVIVED. That was my harness, not your branch. I ran it in a git worktree with a symlinked node_modules, which fails vite's transform step, so every suite "failed" for an unrelated reason and the gate read twelve identical non-failures as twelve regressions. The harness now refuses to run without a green baseline first. Twelve identical results should have told me it was systemic before I read it as a finding.

Confirmed positively

  • serving excluding allowedClients is right, and status-parity.test.ts pinning it against mcpRuntime is the guard that matters — the panel and the endpoint reading the same primitives in the same order is what stops them drifting into disagreeing. Your reasoning in the body is correct: verification is JWT-based and never consults the allowlist, so an empty allowlist stops a new registration and nothing else. Folding it into serving would report a working endpoint as down.
  • The fault string carries no key material. Since it now renders in the UI I checked every unusable() call site: all details are static prose, and the only interpolations are a key count, an errno, and the variable name (plus the path, for the file form). Nothing echoes the value. Worth having checked given where it now appears.
  • MCP_PATH single-sourced and the connector taken from the derived string rather than recomposed — correct, and the reason is exactly as your comment says: the token is audience-bound to that byte sequence.

InstancePanel.svelte's help text now says MCP derives its connector URL and issuer from public_url, and the panel puts a Change link next to it. What neither says is that changing public_url invalidates every issued MCP token.

Tokens are audience-bound per RFC 8707, and createJwtVerifier checks both audience: deps.resource and issuer: deps.issuer. Both are derived from public_url, so editing it makes every existing token fail both checks — 401, and the client must re-discover and re-authorize. The discovery document changes underneath it at the same moment.

That is recoverable, and the same cost as rotating the signing key. But this panel is the thing that will drive operators to that setting — it is the most common reason MCP is not serving, as you note — so it is the one place the consequence should be stated. Something like "changing this disconnects any connected MCP client; they will need to authorize again" on the confirmation or in the help text.

Not a blocker, and arguably it belongs in #170's help text rather than yours — say the word and I will add it there instead, since public_url's MCP coupling is my branch's doing.

Two things I liked

The dead Escape handler is a real bug found only by driving a browser — the handler sat on the dialog while focus stayed on the row's Remove button, so the keydown bubbled up a different branch. document.activeElement was BUTTON.btn sm. That is exactly the class jsdom cannot reach, and spinning it out as a shared Modal.svelte rather than patching one copy is the right call given PassphraseRotationModal has it too.

And reporting that your own "no request was sent" check was worthless — the server logs no requests, so the grep proved nothing — is the kind of correction that is easy to quietly drop. It is the same shape as the absence-grep trap this repo has hit before: a check that returns nothing because it never ran looks identical to one that passed.

Review from the session that built #170, which this is stacked on. I checked the things I am uniquely placed to check — whether #170's guarantees survive your changes — rather than re-reviewing what your own four agents already covered. ## #170's guards all still bite under this branch This PR modifies code #170 owns (`deps.ts`, `jwks.ts`, `singletons.ts`, `server.ts`), including removing a duplicated error literal between the 503 and the panel. That touches the anchors of the mutation gate #170 shipped on, so I re-ran it against `eae9a1d`: ``` baseline: 8 target suites green before mutating KILLED audience KILLED origin-leg KILLED issuer KILLED https-check KILLED typ KILLED key-exclusive KILLED scope-admit KILLED key-single KILLED tool-filter KILLED key-curve KILLED host-order KILLED gate-placement 12/12 guards bite. ``` Audience/issuer/`typ` binding, the tool-surface admission rule, the registry filter, Host-before-bearer ordering, the Origin leg, the HTTPS refusal, both-spellings exclusivity, single-key and EC P-256, and the gate placement that keeps a vault key off `/api/mcp` — all still kill their named tests. Full suite on your head: **195 files / 2190 tests, exit 0**, matching your claim exactly. **A false alarm I caused, in case you saw anything odd:** my first run reported `0/12` — all twelve `SURVIVED`. That was my harness, not your branch. I ran it in a `git worktree` with a symlinked `node_modules`, which fails vite's transform step, so every suite "failed" for an unrelated reason and the gate read twelve identical non-failures as twelve regressions. The harness now refuses to run without a green baseline first. Twelve identical results should have told me it was systemic before I read it as a finding. ## Confirmed positively - **`serving` excluding `allowedClients` is right**, and `status-parity.test.ts` pinning it against `mcpRuntime` is the guard that matters — the panel and the endpoint reading the same primitives in the same order is what stops them drifting into disagreeing. Your reasoning in the body is correct: verification is JWT-based and never consults the allowlist, so an empty allowlist stops a *new* registration and nothing else. Folding it into `serving` would report a working endpoint as down. - **The `fault` string carries no key material.** Since it now renders in the UI I checked every `unusable()` call site: all details are static prose, and the only interpolations are a key *count*, an errno, and the variable name (plus the path, for the file form). Nothing echoes the value. Worth having checked given where it now appears. - `MCP_PATH` single-sourced and the connector taken from the derived string rather than recomposed — correct, and the reason is exactly as your comment says: the token is audience-bound to that byte sequence. ## One finding: the `Change` link invites an action that disconnects every client `InstancePanel.svelte`'s help text now says MCP derives its connector URL and issuer from `public_url`, and the panel puts a `Change` link next to it. What neither says is that **changing `public_url` invalidates every issued MCP token.** Tokens are audience-bound per RFC 8707, and `createJwtVerifier` checks both `audience: deps.resource` and `issuer: deps.issuer`. Both are derived from `public_url`, so editing it makes every existing token fail both checks — 401, and the client must re-discover and re-authorize. The discovery document changes underneath it at the same moment. That is recoverable, and the same cost as rotating the signing key. But this panel is the thing that will *drive* operators to that setting — it is the most common reason MCP is not serving, as you note — so it is the one place the consequence should be stated. Something like "changing this disconnects any connected MCP client; they will need to authorize again" on the confirmation or in the help text. Not a blocker, and arguably it belongs in #170's help text rather than yours — say the word and I will add it there instead, since `public_url`'s MCP coupling is my branch's doing. ## Two things I liked The **dead Escape handler** is a real bug found only by driving a browser — the handler sat on the dialog while focus stayed on the row's Remove button, so the keydown bubbled up a different branch. `document.activeElement` was `BUTTON.btn sm`. That is exactly the class jsdom cannot reach, and spinning it out as a shared `Modal.svelte` rather than patching one copy is the right call given `PassphraseRotationModal` has it too. And reporting that your own "no request was sent" check was worthless — the server logs no requests, so the grep proved nothing — is the kind of correction that is easy to quietly drop. It is the same shape as the absence-grep trap this repo has hit before: a check that returns nothing because it never ran looks identical to one that passed.
lz changed target branch from feat/mcp-resource-server to main 2026-09-17 13:54:33 +02:00
lz merged commit 192f40867b into main 2026-09-17 13:58:52 +02:00
lz deleted branch feat/oauth-settings 2026-09-17 13:58:54 +02:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
lz/agent-nexus!172
No description provided.