The token-meter projections now subtract compact/summary's
shadowedTokenCount verbatim, so the seeded-history fixture's hand-picked
10k count drove the message figure negative and killed the projections
baseline. Realize the seed first (realizeSeedFixture, idempotent, shared
with seedSession) and derive the count from the shadowed rows through
the composed meter, reproducing the previous per-node fold bit for bit —
the recorded aria goldens pass unchanged.
Injected context reached the transcript as one anonymous shape whose
expanded body was the whole message serialized as inline JSON, so the
escaping collapsed the only readable part — the model-facing prose —
into a single line.
`MessageSource` gains an optional producer-declared `form`, a small
semantic vocabulary of information shapes independent of `kind`: kind
says who produced the context, form says what shape it is, so several
producers may share one presentation. Two values ship.
`instructions` (workspace-context) lists the reconciled files above
their text and keeps the `<system-reminder>` framing verbatim, because
the framing is part of what the model read. `catalog` moves dsh-tool-skill
off the shared plugin kind onto a `skill-catalog` source carrying the
published name/description entries, and the body lists those instead of
re-parsing `<available_skills>` out of the prose. Catalog identity moves
with it: the republish digest now covers the durable entries, deleting
the text-slicing that recovered them from a logged message.
Everything else renders the opaque body — the model-facing text with its
real line breaks, then the remaining provenance as fields. That is the
documented default, not a leftover: a resumed, forked, or foreign log
must render whether or not its producer is mounted here, which is why
the classification lives in the durable source rather than a client-side
table keyed by producer.
Review follow-ups on the context provenance row:
- the producer name now follows ToolRow's aria-hidden separator dot instead
of a literal middot inside the text, so the two disclosure rows expose one
accessible-name shape;
- the English steering caption becomes the noun `Interjection`, parallel to
the Chinese 插话 and to the label this bubble carried before;
- the history fold gets its own provenance assertion, so the mapping it
declares separately from TranscriptAdapter is pinned on both sides;
- the superseded-in-part bullet on the no-interjection-chrome note now
enumerates the third Decision bullet too, whose no-badge and no-IconActions
clauses had both lapsed.
Every logged non-user user/message collapsed into one identical
「上下文注入」 row, and mid-turn steering rendered in exactly the bubble a
turn-opening prompt uses, so the transcript could not say what had been
added or which message interrupted a running turn.
TranscriptAdapter and the history fold now attach a provenance view to
each context node, computed by contextProvenance() from the durable
source alone: a role (inject, or recall for a cross-session snapshot)
and a producer name read out of the log — instruction paths for
workspace-instructions, session titles for session-reference, the plugin
id for a plugin source, and the bare kind for any other. No client-side
table of producer names, so a renamed or newly mounted producer stays
identifiable without a client release and a foreign log projects like a
live one. ContextInjectionRow titles itself from the role and shows the
name beside it; MessageItem captions durable and pending steering
bubbles.
The caption reverses one clause of the no-interjection-chrome decision,
which removed it because the composer could not steer; composer steering
shipped afterwards without amending that note, so this change supplies
the product decision its reintroduction clause required and corrects the
stale facts left in it.
Fixes#1291
Steer, inject, and followup now land as durable user/message events on the
session surface; the steering/message event type and its ConversationNode
kind are removed from the client projection. Update tests, docs, generated
catalogs, and agent notes to match, and align the steering e2e fixture and
prompt inventory assertions with the durable user/message landing.
Both Workspace surfaces offered "Open local folder…" and "Create a new
workspace" for one outcome. The browse occupant already carries its own
New folder affordance, so picking a directory covered creating one; the
name dialog only added a second vocabulary and a create target the
operator could neither see nor choose.
The surviving entry is named after the outcome — "Add workspace…" — and a
menu now appears only where there is something to choose between: with no
Workspace listed (the add-only sidebar header, or an empty hero list) the
anchor gesture raises the directory flow directly instead of a one-row
popover. An empty list counts as final only after the list baseline lands,
and a composition with no directory-flow occupant hides the sidebar button
rather than offering a dead one.
WorkspaceCreateFlow becomes WorkspacePickFlow (createOnly -> addOnly) and
the injected createWorkspace narrows to { path }. The host's
workspace.create({ name }) branch and `dsh web --workspace-root` lost their
last product consumer; both are marked at the call site for a follow-up.
Review round: the no-caption rule the Agent Note states applied only to the
success texts, leaving `permission · unknown permission preset "bogus" (…)`.
The error text now reads `unknown preset "bogus" (…)` and its exact wording
is pinned; the fixture mirror drops `JSON.stringify` for the host's own
quoting so the two cannot drift on a quoted argument.
The Note now states the line it draws: the rule bans a caption for the
command's own value, not the vocabulary, so `/plan`'s `Plan mode off.` and
`/goal`'s `Goal cleared.` conform as written — recorded with the broader
name-ban as a rejected alternative. The web row assertions are scoped to the
row so unrelated page text reading `permission` cannot satisfy them.
The web command row renders `title · summary` from one logged command
lifecycle pair, and the two halves were written without knowing about each
other: the title was the dispatched line rebuilt from `command/run` and the
summary was `command/done`'s verbatim text, so every Access-chip pick read
`/permission workspace-write · Permission preset: workspace-write.` — the
command name twice and its argument twice.
The title is now the bare command name (no `/`, no arguments — the summary
already says what the command did), and a command handler's settlement text
never repeats the command's own name, so `/permission` returns `preset
workspace-write`. The row reads `permission · preset workspace-write`, and
the TUI notice still names the preset that now applies. The log is
unchanged: `command/run` keeps its structured name/args split for a richer
registered row.
Move COMPACT_CHECKPOINT_SOURCE and isCompactCheckpointSource into a
cordis-free src/checkpoint.ts leaf, re-exported from the root so every
host-side consumer keeps its import. The client can then type-import the
leaf without reaching dsh-session's root, whose Context merge declares the
host sessions service and collides with the client's -- the dsh-commands/brand
shape. Renaming the plugin id now fails the client typecheck.
Also: keep recoverable summary text when a compact/summary mixes text with
other block types, and capture the seeded-history provenance seqs from the
pushes that produce them instead of deriving them by arithmetic.
Replace the surface-ordered fold with a log-ordered human transcript:
append-origin surface events at their own log positions plus one marker per
landed compaction checkpoint. Command folding, the tool-call index, and the
rev-keyed memo carry over unchanged.
Removes foldDegraded, the padding sentinels, baseSeq, and degradedSeqs() --
they existed only to satisfy the core fold's seq === index assertion. That
also closes the pagination hole A1 exposed: a page can carry a checkpoint
whose shadowed range fell outside the window, and nothing resolves
surfaceOp.start anymore.
The shared test module was named harness.ts inside a repo whose product
IS a harness — hopelessly ambiguous. Renamed to scaffold.ts with
launchWebScaffold/WebScaffold; tsconfig plane-split entries, the seam
JSDoc/README mentions, and both Agent Note languages updated.
Both scenarios gain a Playwright interaction step over the settled
transcript (after the golden capture, so committed aria surfaces stay
untouched): replay-round-trip clicks the reasoning fold open/closed over
wire-delivered state; seeded-history expands a read tool row rebuilt from
the cold log and asserts the recorded result text appears (read rows are
expand-in-place — rowExpands routes the click to the inline fold, not the
details column). test:web 30 passed | 1 skipped.
apps/web/tests/harness.ts boots the real web assembly in-process
(startHost llm:false -> installLlmReplay providers-mode -> mountWebPlugins
-> startWebServer) under DSH_SNAPSHOT replay/record/refresh. Barrier
stack: in-process turn/end -> agent.whenIdle (covers the persistence
flush) -> browser settled-poll. Seeding goes through the real persistence
API (semantic-checkpoint precedent); record harvests fixtures from live
session memory and tokenizes {{sessionId}}/{{cwd}}; refresh is the sole
golden writer. Console tripwires fail scenarios on reconnect/gap-repair
self-healing; harness close asserts full replay-fixture consumption.
Scenarios, each with fixtures recorded against THIS assembly via a live
model run: replay-round-trip (real composer -> real bash echo -> settled
markdown + aria golden + world-state event asserts) and seeded-history
(cold sidebar list -> implicit resume on open -> history tool cards from
the log, zero model calls). apps/web/tests are host-plane programs:
excluded from the client-registered apps/web project, included in
tsconfig.host.json (one program cannot hold both Context merge sides).
