The remaining P1 from the #939 review, plus the P2 it shares a mechanism with.
Nothing carried a version, so two tabs editing one namespace silently
overwrote each other — reproduced as tab B's `reasoning` lost to tab A's
older draft. The seam's per-namespace write queue orders writes; it cannot
tell a fresh writer from one replaying a snapshot a predecessor superseded.
Each namespace now carries a monotonic `revision` over its RAW section. A
write may send `expectedRevision`, checked at the FRONT of the queue (not at
call time, which would race the very predecessor it guards against); a
mismatch rejects with `SettingsConflictError` → `settings-conflict` on the
wire, carrying both revisions. The editor captures the revision it opened at
and, on conflict, asks the user to reopen rather than replaying its snapshot.
The same counter fixes the missing broadcast. `settings/updated` is gated on
the resolved value — correct for consumers, wrong for configuration surfaces:
storing an override equal to the composition base leaves the resolved value
alone while changing what the document says (the field is now overridden, not
inherited) and moving every open editor's revision. `settings/document-updated
(ns, revision)` fires on any raw-section change, in-process or external, and
`host/settings-changed` now rides it.
That event also closes the stale model picker: editing a provider's `models`
changes no route, so `llm/adapters-updated` never fired and an open picker
kept serving the old catalog. A change to an exposed provider namespace now
emits `host/models-changed` too — that namespace holds the catalog.
Docs: both sides of the five touched README pairs, a type-equiv block for
`SettingsPathOp`, and an Agent Note recording what the plane exposes and who
may overwrite what. The deferred wire-redaction gaps (secrets behind
union/intersection/transform, `.default(...)` in the served envelope, schema
text in rejection messages, `new Function` rehydration, pi-ai's `headers`) are
recorded as TODO(settings-wire-redaction) and in Known Limitations rather than
half-fixed.
Five findings from the #939 review, each reproduced before being fixed.
**Configuration reads are as privileged as writes.** `settings.describe`
returns every exposed namespace's configuration and `credentials.describe`
reports whether an arbitrary environment-variable name is configured and from
where — reconnaissance no anonymous caller should have. Both join
PRIVILEGED_METHODS, so the whole configuration plane is loopback-only until
real authentication exists; `trustedHosts` was never authentication. The model
catalog stays reachable: it carries no endpoints or key state, and a LAN
client's model picker legitimately needs it. Asserted over a real HTTP server,
because the Host header a browser actually sends is what decides this.
**The proxy serves only namespaces a registered model provider addresses.**
The settings seam is general — any plugin may register one — but the Web
configuration plane is the model-provider surface. Without the gate, every
future `settings.register()` would silently become remotely readable and
writable configuration. An unregistered namespace and an unexposed one answer
identically, so no caller can enumerate the registry one probe at a time.
**Path-addressed writes replace the redacted-document rebuild.** The editor
reads the REDACTED descriptor, so rebuilding a section from it and replacing
wholesale deleted every literal secret the wire never returned — reproduced as
`{baseURL, reasoning}` in, stored `apiKey` gone out. `settings.mutate` applies
set/unset ops to the section as it stands at the front of the seam's write
queue, and the client names only fields it can see, so an unseen secret is
untouched by construction rather than by care.
P2s in the same pass: `llm/adapters-updated` now contains async listener
rejections (an uncontained one escaped as unhandledRejection, contradicting
the documented "observer failures are contained"); llm-deepseek's retry-policy
swap uses the atomic `registration.replace` instead of dispose-then-register,
which published `[]` then `["deepseek-official"]` so an observer saw the
provider disappear and come back; and a transport rejection no longer strands
the page in `loading` or a card in `busy`, with removal failures surfaced on
the page banner instead of swallowed.
- `SessionSummary.updatedAt`'s wire doc still said "Persisted file mtime",
which stopped being true for attached sessions.
- The core invariant let `session/inherited` fall through the merge-extensible
default. It is core-owned, so it gets an explicit case; an unbalanced seed
legally places it inside an open turn, which the relation permits.
- The Agent Note claimed the boundary reaches disk via `live.pending`/
`scheduleDrain`. Verified false: the constructor append precedes `enter()`,
so it never publishes on `session/event` and rides the creation seed instead.
Attaching is therefore a write where none happened before — recorded, since
only `load()` stays a pure read.
- The deferred-index proposal asserted this change documented the cold-mtime
skew on `dsh-host-apiproxy`. It did not; the README entry now exists.
- `firstLiveSeq`'s firehose gap runs through its own seq, not below it.
- The boundary is not always at `firstLiveSeq` (the idempotence guard), so
consumers scan for the last one.
- `lastActivityTime` excludes by type, so a pickup time still leaks onto a
synthetic closer when a boundary ends an open turn. Documented.
- Pin the fork claim end-to-end: a child inherits a still-running parent's
open bracket below its own boundary, while the parent has none. Fails if the
write moves back to the load path.
- Fix the telemetry title that contradicted its own assertions.
The `/status` call site cannot be pinned the way the other two are: the
command appends its own `command/run` before rendering, so the boundary is
never the log tail there. Its fixture now at least renders over a
boundary-bearing log.
A plugin owning a standalone open/close bracket cannot tell a dead marker
from a live one: an unmatched `compact/start` reads identically whether the
previous writer died mid-compaction or a compaction is running now.
`Session.firstLiveSeq` already holds that answer exactly, but only in memory.
Append the log-only `session/inherited` event at that seq from the seeded
constructor — the single waist all six seeded-start paths pass through
(resume, configured startup on a persisted id, `sessions.fork()`, a subagent
fork child, `adopt()`'s live prefix, and a bare seeded `create`). Read it
through the new `isInheritedSeq(events, seq)`.
The constructor placement means persistence needs no changes: the marker is
already in `events` when a backend captures the creation seed, so it rides
the ordinary seed path with no load-time write. It also covers fork, where
the inherited bracket's owner may still be running — the case a
persistence-layer boundary could not reach.
Activity ordering excludes the boundary through `lastActivityTime()`, since
lazy resume makes browsing a pickup and the three call sites would otherwise
float every opened session to the top of a picker or list.
Eight compiler-locked methods: settings.describe/update/replace serve
redacted layered namespace views (secrets structurally absent from every
layer, write-only in the update direction) and fold seam refusals into
settings-rejected; credentials.describe/set/unset expose value-free views
with credential-rejected on shadowed writes; llm.providers merges the
configurable directory with live routes and llm.models claims the
host-scoped catalog reservation through the buildModelCatalog extraction
session.models now shares. Three HostFrame invalidations bridge the seam
events (host/settings-changed, host/credentials-changed,
host/models-changed), and the connection route generalizes the native-
dialog check into a privileged-method set covering all four writes. The
fixture and both fake clients grow the same face.
The rename impl narrows on SessionTitleInvalidError: only an
empty-normalizing title maps to title-invalid (its message renders verbatim
in the rename dialog alert), while liveness/disposal races fall to internal;
the absent-service message trims to one presentable sentence. rpc-schemas
gains the title-invalid accept/missing-details lines; cosmetic ordering
(type-only import comment, tsconfig reference, schema import order) restored.
New unary method in RpcMethodMap with the title-invalid error code; the impl
resolves the agent (cold sessions resume first) and delegates to
ctx.sessionTitle.rename, returning the normalized title plus its event seq so
clients settle the title projection cell ahead of the push frame. session.fork
stays on the reserved-seam list.
The native adapter's route was named deepseek, colliding with pi-ai's
catalog provider of the same name, so the two DeepSeek paths could never
be mounted side by side. The web settings page needs both configurable at
once. Compositions, fixtures, goldens, scaffolding defaults, and docs all
move together (pre-release, no shim); TUI/session-query-spill/
missing-credential goldens re-recorded through their keyless refresh
modes because provider-name length shifts box padding and spill
truncation points.
Two races from the #572 review, still live in the ported registry:
An ask whose signal aborted between the service's own check and the
microtask-deferred waterfall dispatch would register its abort listener
AFTER the signal fired — never invoked, entry pending forever, zombie
frame on every mux replay. The answerer now settles 'cancelled'
synchronously before publishing anything.
The audit back-scan let a callId-less ask claim the newest unclaimed
asked record even when that record carried another call's id. Pairing is
now shape-symmetric: callId-bearing asks take exactly their call's
record, callId-less asks take only callId-less records — neither can
steal under parallel asks.
Disposability parity with the question provider: a gateway disposed while
approvals are pending settles every registry entry as 'cancelled' (the
service's fail-closed vocabulary), so no ctx.approval ask dangles past the
proxy's lifetime and mux subscribers see the withdrawal. Spec mounts the
proxy on its own fiber and drives dispose with a live ask.
Addresses the ds-review-bot suggestion on PR #851.
Selecting a preset from the hero pushed the session into the conversation
view: the /permission switch logs its command/run + command/done pair, the
pair folds into flow nodes, and the composerPhase predicate counted ANY
node as conversation — so the hero (composerPhase === 'blank') collapsed.
The host-side blank bit was already correct (sessionBlank = no turn/start;
knob events open no turn), but the client derives its phase from window
content, and command rows are log-only records, not conversation.
derivePhase's hasContent now excludes command nodes — the client mirror of
the host predicate. The knob events themselves never fold (not
surface-eligible), so the pair was the only leak. Covers /plan on the hero
identically (same lifecycle pair, same predicate).
Specs: the host blank spec pins the three knob events as standalone
events; a session spec drives the /permission pair through the live path
and asserts phase stays 'blank' while the command node renders.
Every filesystem await in the browse scan (opendir and each read) now
races the signal through raceAbort, so a stalled network open/read stops
with a departed caller and an already-aborted request rejects even for
an empty level; the abandoned settlement is swallowed and an abandoned
open that still mints a handle is closed, never leaked. apiproxy maps an
aborted listing to the cancelled wire code, matching pickDirectory and
command.execute, instead of reporting a false internal failure. The
fixture spec call sites gain the wire signal argument the previous
commit's static lane flagged.
capability.list gains an optional AbortSignal threaded from the RPC
carrier's request signal (the pickDirectory pattern): a disconnected or
timed-out caller stops the opendir loop instead of the scan outliving
its caller, and the abort surfaces as its own reason rather than a
directory-unreadable dressing. boundedInsert rejects a full window's
at-or-beyond-tail candidate on one comparison and binary-inserts
retained candidates, so an oversized level no longer pays a window scan
per dirent.
Shared-surface conflicts resolve as unions: the fixture serves all five
projection keys (title/todos/permissions/plan/goal) with the /permission
and /plan command mirrors side by side, the connection specs assert the
five-key baseline and the shifted approval/question replay indices, and
the cli roster/deps, tsconfig aggregate, and README allowlist carry both
lines' rows. Plan-side content lands verbatim from master.
One end-state consolidation both branches half-did: with questions
(ui-question) and approvals (ApprovalPanel) each owning a composer
takeover, PendingCard retires outright — ChatView renders no pending
placeholder, the card component and its specs go, and both README halves
state the takeover-only contract.
One list call now materializes at most maxEntries child rows (config,
default 1000 - GitHub's web-UI directory-listing bound). Candidates sort
before probing so a cut level keeps the name-sorted head and symlink
probing stops with the bound, and DirectoryListing carries a required
truncated flag on the seam and the wire so clients can state
incompleteness instead of silently missing tail entries.
Union resolutions throughout — the fixture serves both the goal and plan
projection units (catalog gains /goal beside /plan; the retired
goal-fixture sample command yields to the real goal mirror), the mux
baseline spec expects all four unit frames, and the tsconfig paths /
Model Experience allowlist carry both domains' outlets.
