The terminal and history pagination both treated the model-visible surface as
the human transcript. A landed compaction replacement therefore erased the
conversation it summarized — messages the reader had already seen — and a
model-only replacement copy consumed a page's `maxMessages` quota, which could
also split a compaction's provenance from the replacement citing it.
`dsh-session` now exports the marker split `isAppendSurfaceEvent` /
`isReplacementSurfaceEvent`. The terminal replays append-origin surface events,
keeps a shadowed step's tool cards paired through its append-origin assistant
message, and renders one dim marker where a compaction landed; the checkpoint is
recognized through the compaction seam's `isCompactCheckpointSource` contract,
not the shape of the replacement. `session.history` counts only append-origin
human messages. Everything model-facing keeps reading `session.surface`.
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.
Shared-surface conflicts resolve as unions: the fixture serves all four
projection keys (title/todos/permissions/goal) with both the /permission
and /goal command mirrors (the goal-fixture placeholder retires with
master), apps/cli carries both lines' dependency additions, and the README
Model Experience allowlist keeps both entries. The connection specs assert
the four-key baseline and the shifted approval/question replay indices;
the module graph regenerates over the merged dependency set.
Three ui-conversation spec conflicts resolve to master's SlotTestRuntime
rewrites. Adaptation to the new outward session face: ISession gains the
command verb (the composer chip and the /permission picker submit through
it), FixtureSession grows the matching fail-loud stub plus the
waitingApproval summary default, and the picker reads the projection
through projections.faceOf (the ProjectionsFace shape) instead of the
retired store getter.
The summary blank bit switches from log emptiness to the absence of any
turn/start: standalone plugin events — command lifecycle records,
plan/mode, session titles, goal metadata — no longer surface a fresh
session in lists or steal the New Session view. Running /plan (or /goal)
on a blank session keeps it blank and reusable; the first accepted
prompt's turn clears it. Both carriers share one predicate (summarize +
the host/session-added frame); the cold path keeps its constant false
with the index-read rationale; the client mirror already flips only on
prompt acceptance and needed no change.
Review finding (PR #791): the column carried bare values (no seq), so the
client could not seed its value store without risking a stale list block
outranking newer push frames — and nothing consumed the column at all,
leaving cold titles absent after a restart. SessionSummary.projections is
now the same SessionProjectionsBlock as the history tail (values +
asOfSeq; attached rows cut the live registry, cold rows serve the cache's
identity-checked cachedSnapshot whose asOfSeq is the lowest served-row
watermark). SessionManager.refreshList seeds each row's block into the
per-session projection store via per-key apply — partial-baseline
semantics: an absent key never clears, and higher-seq-wins keeps stale
list blocks beneath push frames and tail baselines — so cold titles
surface in the sidebar without opening a session.
SessionSummary grows an optional projections column (whole value per key,
same passthrough posture as the history-tail block): attached rows cut the
live registry watermark cache; cold rows view the persisted projection
cache's stored rows via the new registry viewCheckpoint face (version-
matching keys only, zero I/O) — the RFC's motivating scenario, every
session's title across a listing without loading one event log. The column
is fail-soft and absence-coded: no registry, no cache row, or a throwing
read serve the row without the column, never breaking the listing.
ui-workspace's two trigger surfaces each declare a single-kind directory-flow
hole (conversation.hero.workspace.directoryFlow / sidebar.workspaces.directoryFlow,
same owner contract) and keep only the trigger and the adoption: the Open-local-
folder entry renders while the surface's hole is occupied, and the occupant
reports one picked path per open through the hole's owner conversation
(open/busy/onPicked/onCancel/onError).
directory-picker-native becomes dual-face: its browser half fills both holes
with a renderless occupant driving host.pickDirectory, so the cordis.yml row
that mounts the backend also composes the client interaction — a mismatch is
impossible and a second flow package fails at client load.
With composition wiring both sides, the host.describe.directoryPicker
advertisement and the client's kind branching lose their last consumer:
the field, WorkspacesService.directoryPickerKind(), the DirectoryPickerKind
wire type, and the picker's per-open describe read are deleted. The connection
fixture now serves a deterministic pickDirectory path so the keyless snapshot
drives the full pick-then-adopt flow. ui-workspace's hand-rolled declaration
deferral is replaced by the deferRegistration helper it duplicated.
The approval/request waterfall answerer fills the slot the gateway reserved
('registry absent in this minimal version'): an ask through ctx.approval
becomes an answerable approval/requested mux frame with a stable rpcId,
mux open replays still-pending frames (refresh recovery), respond routes
approvals first by the echoed rpcId and cross-checks the payload's audit
correlation, the ask's abort signal withdraws with a broadcast cancelled,
and approval/resolved settles every subscriber. The answerer pairs each ask
with its approval/asked audit event by scanning for the newest undecided,
unclaimed id (callId-matched when the ask names a call); asks that bypassed
the audit path delegate to the fail-closed default.
The child activates only when ctx.approval is composed; the 275-line spec
carries over verbatim at the gateway's new home.
The session.permissions/setPermission unary pair, the PermissionOption wire
DTO, the client Session wrappers, and the fixture/fake mirrors all leave the
wire: the read side moves to the 'permissions' session projection and the
write side moves to the /permission command in follow-up commits, so the
web protocol gains no permission methods at all.
The pendingSwitches + prompt-submit flush + hasOpenTurn move also goes.
Knob events no longer need turn enclosure: the persistence scanner keeps
standalone events after the last turn/end as part of the preserved prefix
(remove-synthetic-log-only-turns), none of the three knob invariants demand
an open turn, and the setters append bare events. An idle switch commits
immediately; hasOpenTurn stays a user-approval private fold (its audit pair
is the one contract that still requires enclosure).
The old PermissionSelect chip and its mount-time fetch die with the RPCs
(the resident composer broke the mount-once assumption); the projection-fed
replacement lands with the Access seat swap.
master's toolcall-open extracted runNativeCommand inside apiproxy for the
openPath opener while the picker seam had moved the native chooser (its other
consumer) into directory-picker-native; after the merge the two packages each
carried a verbatim copy. The runner now lives in packages/util/native-command
(zero-dependency library, per the util-group contract) and both native
integrations depend on it.
39 conflicts resolved per the reattachment plan (missions worktree-projbiz
0728-1859): baseline wins for deleted packages (host/runtime, old ui/acp,
ui-sidebar Rows/tree) and retired specs; unions for wire-layer exports and
client summary fields; the approval takeover, waitingApprovals tracking, and
PendingApproval domain face carry over onto the master structure. The two
new host specs follow the runtime->apiproxy rename. Dead PR-side wiring
(ConversationInjected permissions/setPermission spread, InputBar controls
prop, boot.ts sandbox composition) resolves to master and its replacement
lands in follow-up commits.
The browse interaction also presents a dialog (the in-app modal), so 'dialog'
failed to discriminate the two capability kinds; 'native' names where the
chooser runs. Package directory-picker-dialog -> directory-picker-native, kind
'dialog' -> 'native', with every seam/gateway/client/doc reference updated and
the seam Agent Note's naming rationale rewritten to match.
Six mutation RPCs (create/edit/pause/resume/complete/clear) move into
dsh-host-apiproxy (the PR's host/runtime carrier is gone): goalService()
resolves ctx.get('goals') with a loud absence error, mutateGoal() resolves
the session's agent (agentFor, the command.* implicit-resume precedent) and
acknowledges with the new CAS ref only. GoalError codes ride err.details.
goal.get and the wire GoalView/goalViewSchema are gone: the read side is the
'goal' session projection (whole values on the history tail page and
session/projection frames), so responses never feed client state — the rule
whose absence forced the original PR's write-revision fences.
Conflict rulings follow the projection-reattach plan:
- host/runtime package (deleted on master): take master; the PR's boot
composition moves to the cordis.yml roster and its goals handlers will be
re-landed in dsh-host-apiproxy; the session.prompt slash interception and
its spec are dropped entirely (superseded by command.execute + command/run
logging).
- client core (rewritten on master): take master; the PR's Session goal
fields/methods, ConversationSnapshot.goal, goalActions injection, and the
hard-mounted GoalBar are all superseded by the 'goal' session projection
(useProjection) and will return as the ui-goal plugin.
- wire contract: union of master's workspace/command/skill domains and the
PR's goal domain, minus goal.get (the read side is the projection block +
session/projection frames; six mutation RPCs stay).
- GoalBar component and spec leave ui-conversation (they re-land in the new
ui-goal package); IconSparkle16 stays in ui-conversation chat.
- The web-slash-command-dispatch note documents the dropped interception and
is removed; the goal-bar note will be rewritten for the projection model.
- pnpm-lock.yaml taken from master (reinstall recomputes).
Sweep the silently auto-merged leftovers of the superseded #587 wire layer:
the PlanModeState interface and planMode/setPlanMode entries in the sessions
API and rpc map, the fixture plan RPC spec, the composer.controls README
paragraph and spec assertions, the old-format plan-mode snapshot, and the
two Agent Notes written against the RPC design (to be rewritten for the
projection reattachment). ChatView/PendingCard keep the approval-only
pending rendering; QuestionComposer improvements ride along unchanged.
Resolution follows the reattachment plan: the #587 wire layer (planMode/
setPlanMode RPC, prompt.planMode, client Session projection fences) is
dropped in favor of the session-projection base now on master; master
structure wins in all shared files. Kept from the PR side: the approval-only
pending filter in ChatView (questions render solely through the composer
takeover) and the auto-merged QuestionComposer improvements. The deleted
host/runtime package and retired test carriers are removed with master.
ds-review-bot round 1: the seam documented a merge-extensible union but
shipped a closed alias, and the gateway schema rejected any kind beyond
dialog/browse — a third backend could neither implement the seam nor be
advertised. The union now derives from an augmentable DirectoryPickerCapabilities
map, host.describe.directoryPicker preserves unknown wire kinds, and the
browse fixture applies listDirectory's root special case so creating under
'/' no longer mints a '//name' identity.
The full-suite coverage gate found the new listDirectory/createDirectory
client methods and handler routes unexecuted: the implementation and
schema layers were tested directly, but nothing crossed the wire form.
One round trip through InProcessApiClient covers both arrows on each
side.
The web GUI's folder picking was hardwired to one interaction: a native
OS chooser compiled into the gateway, unusable for remote deployments
and swappable only by editing apiproxy source.
Directory picking becomes a three-package capability seam in
packages/host: ctx.directoryPicker returns a discriminated capability —
dialog (the extracted native chooser; host-display only) or browse
(new: one-level listing + child creation over Node stdlib, hidden flags
host-stamped, symlinks followed, ancestry crumbs; remote-capable). The
gateway injects the seam, advertises the kind via
host.describe.directoryPicker, serves host.listDirectory /
host.createDirectory under browse, and answers
directory-picker-unavailable across kinds. cordis.yml is the swap
point; apps/cli keeps dialog mounted, so behavior is unchanged until
the in-app browser PR flips the default. The connection fixture serves
a deterministic browse tree; WorkspacesService gains the browse calls
the browser UI will drive. Decision record:
.agents/notes/implemented/architecture/2026-07-28-directory-picker-capability-seam.md
The only browser-trust guard covered host.pickDirectory, while the
consequential methods (session.prompt drives bash) accepted any Host —
open to DNS rebinding, where a rebound page reads and writes the API as
if same-origin and only the Host header betrays the attacker's domain.
The pickDirectory-specific loopback guard becomes a prefix-wide fence:
Host must be loopback or an exact host[:port] from the new trustedHosts
config, an attached Origin must equal that authority, and explicit
cross-site markers are refused; requests without browser markers (curl,
tests, native clients) pass, because without a browser there is no
confused deputy. The loopback-socket check is dropped — binding policy
expresses reachability, and the fence is not an auth layer. The Agent
Note records the full threat model and the alternatives.
Browsers send "simple" POSTs (text/plain, form encodings) without a CORS
preflight, so a malicious page could execute side-effectful RPCs blind —
the response stays unreadable cross-origin, but session.prompt would
still run. The carrier now answers 415 unless the declared media type is
application/json, forcing every cross-site attempt into a preflight this
server never answers. Raw-fetch specs gain the header; a new handler case
proves the fence rejects before the impl runs.
