Expanding a stage section drew its rows stacked at the section's origin:
CollapsibleSectionBody passed the section's several field composables straight
into AnimatedVisibility, which lays out a single child, so the rows overlapped at
the top of the section and only the last was visible. The shared body now wraps
its content in one full-width Column, so any multi-row section stacks its rows
under its header.
Every stage section now opens as its own pane rather than growing the stage
screen inline:
- PaneState: add the Section pane level (MAX_PANE_DEPTH = 3) and a saved
sectionKey primitive, with backStep/popTo/encode/decode handling so one Back or
peek pops exactly one level and the open section survives rotation and process
death.
- StageDetailPane: the stage screen keeps a compact collapsed list of its
sections (Artifacts, Output, Input, This run, Stage definition); opening one
shows it as a full-window surface over the stage on a narrow window, or a panel
to the right of the stage pane on a wide one, with its own Scaffold/TopAppBar
title, its own scroll and its own top spacing.
- SectionHeader: a navigation mode for those rows - a static forward chevron
(mirrored for RTL) and an "Open <section>" action instead of a rotating
disclosure chevron and "Expand/Collapse", so assistive tech is not told the row
expands in place. The detail screens' disclosure rows are unchanged.
- DetailTabs: drop the rememberSectionExpansion initiallyExpanded seed, whose
last caller was the inline stage definition.
- Tests: cover the Section level's backStep, popTo, encode/decode and depth, and
the shared stageSectionSpecs order and blank-section cases.
Sections still start collapsed, the operator's choices still persist, and each
section's content is unchanged. android/ only; no dependency was added.
Also in this diff, from the covered-identity statistics work: PipelinesApp passes
compactStats = true on its two stacks (SlidingPaneStack's statistics flag).
The committer staged only the implementer's self-declared changed_files,
so an incomplete declaration silently left part of the approved change
uncommitted. The runner already records the run's real changed paths by
comparing the working tree around each step; publish that set to the
committer.
- runState gains an unbounded, deduplicated, first-seen-ordered captured
path accumulator (capturedPaths/capturedSeen), distinct from the bounded
changedPaths sample used only by the UI-surface skip reason.
- recordChanges folds each captured FileChange path into it in the existing
single pass, so no second walk and no new I/O.
- publishServerChangedFiles writes the set into the input memory value as
input.server_changed_files, called before memoryInput so the binding
always resolves (an empty list when nothing was captured) and only stages
that select it receive it. rebuildRunState rebuilds it from persisted
completed steps, so a resumed run hands the committer the same evidence.
- Tests prove the committer receives the captured set distinct from and
additive to the implementer's declared changed_files, that an omitted
path is still present, and that the field is an empty list when nothing
was captured.
- control.md and orchestration.md document the reserved field and instruct
the committer to select input.server_changed_files alongside the declared
changed_files and stage their union.
Commands run: go build ./...; go vet ./...; go test -count=1
./internal/server/... ./internal/stage/...; go test -count=1 ./internal/author/...;
go test -count=1 -run TestCommitterReceivesServerCapturedChangedPaths ./internal/server/.
All passed. go.mod/go.sum unchanged.
A planner had no machine-readable view of the tool set later stages would
have, so it could emit acceptance checks that required tools (shell) the
implementer was not permitted, causing a blocked run.
modelPrompt now emits exactly one authoritative line naming the stage's
permitted tools, derived from the definition via the new effectiveTools
helper. effectiveTools reuses ToolNames: a present tools block, including
an explicit empty one, is returned as parsed; an absent block yields the
legacy mutating read/glob/grep/edit/shell bundle so it is never
misrepresented as empty or read-only. The bundle is declared once as
legacyTools and reused by tools.go, agent.go, pi.go and permissions.go.
The redundant client-side strings.Replace of the tools line is removed
from AgentClient.Generate and PiClient.Generate; the agent-only guidance
sentence is preserved by appending it in agent.go, and Pi appends a short
mapping line derived from piTools. Edit-scope and permitted-shell text,
routing, review, budgets and retries are unchanged.
Adds internal/stage/model_prompt_test.go covering explicit, absent and
explicit-empty blocks plus derivation edge cases, and a guard that the
agent prompt still carries the tools, edit-scope and shell-command text.
A stage declaring a `skipped` outcome is short-circuited with no model call
when the run's recorded file changes and artifacts touch no UI surface
(android/, internal/tui/, a presentation extension or an image artifact).
The step is recorded completed/skipped along its declared edge with a
bounded one-line reason, so a backend-only run no longer pays for a
ui_reviewer generation that can only answer ui_applicable=false.
A covered pane's width is spent on the covered region (`identityWidth +
startInset`), and when a second level is covered `coveredEdgesDp` gives the
nearest identity that region less the thin older edge (`startInset + minTouch`;
the leading gesture inset cancels), i.e. `identityWidth - minTouch`. Flooring a
statistics-bearing identity at the full block alone (`MIN_COVERED_PANE_WIDTH_DP +
PEEK_WIDTH_DP`, 168dp) therefore left the nearest edge at 120dp, below the
`paneStatTier` Full boundary, so a pipeline's last-run/age and average lines were
filtered out at every depth a pipeline is ever covered. `coveredIdentityWidthDp`
now floors a statistics-bearing identity one touch target above the block
(`+ minTouch`), so the nearest edge is exactly 168dp and reaches Full, while the
single-covered and title-only cases are unchanged and the active pane width still
does not depend on depth.
The covered title also broke mid-word in a narrow strip ("Complete a Task" ->
"Complet"/"e a Task"). `wholeWordTitle` now packs the title greedily onto
whole-word lines and, once they are full, drops the remaining words and
ellipsizes the last line at a word boundary; a word wider than the whole strip is
the one exception, ellipsized to the width because it has no internal boundary.
It takes an injected `measure`, so the packing rule is unit-testable without a
device, and `PaneIdentity` measures with the same `cardTitle` role it draws with.
`PaneIdentityTest` adds the pure cases: the width-degradation tiers, the
statistics-bearing width floor, the nearest-edge geometry at depth 2 (which fails
without the `+ minTouch` reserve, so it pins the defect), and the whole-word
wrap/ellipsis/oversized-word/no-room cases. `android/README.md` records the
corrected geometry, the fix and the run's build/test evidence.
android/ only; no new dependency, no server field, and no server/Go/installer/
Docker change. Build and unit tests only (no device): 300 tests, 0 failures.
The soft-input comments in the manifest, PipelinesApp.kt and android/README.md
still described `adjustNothing`, a mode the activity no longer declares. MainActivity
uses `android:windowSoftInputMode="adjustResize"` and the app draws edge-to-edge, so
the Compose window inset (WindowInsets.safeDrawing) is the single source of the IME
geometry and is applied once by the surface that owns it.
- AndroidManifest.xml: rewrite the soft-input comment to describe adjustResize, why
the app is edge-to-edge and reads the IME from Compose window insets, and warn
against changing it without re-checking the chat composer.
- PipelinesApp.kt: correct the bottom search column comment that claimed the activity
declares adjustNothing and the window is never resized for the IME.
- android/README.md: correct the omni-search results-panel section that claimed the
IME inset was counted once because of adjustNothing; now records the declared
adjustResize mode, the edge-to-edge single source of IME geometry, and the chat
composer's imePadding.
Documentation only; no rendered surface or behaviour changes. The omni-search entry
point and pill morph already exist at this HEAD; this reconciles the stale wording
with the committed mode.
Integrates eight commits of reviewed chat work: the Chat destination with
several conversations, markdown transcripts, tool-call chips with an expandable
INPUT/OUTPUT panel, edit diff cards, the composer's stop control, and approval
cards. The automated resolver failed on this merge repeatedly (its own stages kept
dying on the malformed-result defect), so it was integrated by hand.
Integration decisions:
- Pill geometry: main's tokens win. lane-app2 carried private duplicates of
DESTINATION_ITEM_SIZE/GAP and a private DESTINATION_PILL_RESERVE, added before the
geometry-token migration; those duplicates are dropped in favour of the theme
tokens main already exports.
- IME mode: kept lane-app2's adjustResize for the window, because the chat composer
and the search field both raise the keyboard and the chat UI was reviewed with it.
- Protocol and view-model constants: both sides' independent additions are kept.
- ChatScreen now passes the compactLabel main's SlidingPaneStack gained, and the
literals lane-app2 introduced before the token discipline are converted to five
new named tokens (HAIRLINE_DP, ICON_BUTTON_SMALL_DP, TOOL_PANEL_MAX_WIDTH_DP,
CHAT_TITLE_MAX_WIDTH_DP, CHAT_BUBBLE_MAX_WIDTH_DP) rather than allow-listed.
Verified: assembleDebug plus the full Android unit suite (409 tests) pass,
including the geometry enforcement test.
Integrates nine commits of reviewed work from the lane-go branch after the
automated resolver failed on this merge five times (each attempt dying on the
malformed-result defect whose fix is contained in this same merge).
Integration decisions:
- Search: kept main's shipped design (Store.Search + SearchResponse.Results) and
dropped lane-go's parallel, older attempt (an unused Message.Matches field, an
absent runs.search, and two partial TUI plumbing blocks). lane-go itself did not
compile: protocol.SearchMatch was undefined and model.go referenced searchNav,
requestSearch and applyRunNav, which exist nowhere. Those partial TUI blocks are
removed; the queued TUI omni-search item will implement the surface properly on
main's design.
- Both sides' additions are kept where they are independent: run titles and
provenance, the human decision stage and its CLI/TUI surface, the derived stage
verdict, the stale-client guard, the line-oriented result format, and main's
search command.
- The search key's help label is now "/ omni search"; the two tests asserting the
old label are updated rather than the label reverted.
- docs/cli.md and the packaged operations reference carry both sides' text.
Verified: go build ./... , go vet ./... and the full go test ./... suite pass.
Add references/work.md: one durable ledger per project (queue, active, done,
failed), the start/pause/resume/status operations, per-item directives, a
restart-safe queue driver, sleep autonomy and a conciseness rule. Index it from
SKILL.md and add it to the installed inventory.
Add a "Omni-search results panel (16 Oct 2026)" section to android/README.md
describing the panel that sits above the omni-search field: its placement above
the input and IME, the badge/title/secondary row layout with a bold+underlined
snippet span, the debounced query with stale-response gating and no flicker to
empty, the loading/empty/error states, and result-to-target navigation for every
kind. Records the verification evidence (unit tests and assembleDebug, APK hash
and size, published APK and LAN URL) and notes what was left to the visual review.
No source behaviour changes: the panel logic already existed; this pass verified
it and filled the documentation gap.
Introduce a short, markdown-ish result format whose first non-empty line is
the declared outcome and whose "# output" section carries the routing data,
so stage output stops carrying multi-kilobyte prose inside a JSON envelope.
- New internal/stage/line_result.go: resultFormat() gates legacy JSON ('{'
first) vs the new line format; decodeLineResult() parses outcome-from-first
line, requires a JSON-object "# output" section, stores extra sections
verbatim under normalised keys, and records a reserved result_parse status
(parsed/partial/failed). Failures return a salvage Result plus
ErrInvalidModelResult, and never route an undeclared outcome.
- internal/stage/model.go: write the raw response to raw-output.md (0600,
best-effort) before parsing, and route the line format through
decodeLineResult while leaving the JSON path (stageResultText ->
decodeModelResult -> repairModelResult) unchanged.
- internal/stage/agent.go: agentResultText returns line-format text verbatim
so the "# output" body is not stripped by the JSON-object extractor.
- internal/server/runner.go: merge a non-empty salvage Result into the error
envelope under "salvage" on failure (a no-op for the legacy JSON path).
- internal/stage/line_result_test.go: raw-text fixtures covering happy path,
missing/undeclared outcome, missing "# output", best-effort extras,
adversarial prose, legacy JSON unchanged, >=64KB responses, and
raw-output.md written before parsing.
Combine lane-e's reviewed Android UI work with main's lane-c/lane-d work.
Conflicts resolved by keeping both intents:
- android/README.md: both documentation sections retained
- graph/StageDetailPane.kt: lane-e pane plus local section changes
- screens/RunDetailScreen.kt, RunHistoryScreen.kt: lane-e features and
local work both preserved
- screens/PipelineListScreen.kt: new row layout and remembered
sort/filter controls kept together
Combine the local list-scroll/covered-edge and hoisted pane wiring work with
lane-d's Section pane level, sectionKey persistence, three-way back
descriptions and single-Column CollapsibleSectionBody. Both intents survive:
conflicts in README, PipelinesApp, StageDetailPane, PaneState, Sections and
their tests were resolved by composing both sides, without conflict markers.
Build: ./gradlew :app:assembleDebug
Tests: ./gradlew :app:testDebugUnitTest (322 tests, 0 failures)
Combine lane-c's incoming work with the local branch.
- PipelinesApp.kt was the only unmerged path: resolution is purely additive over HEAD, keeping HEAD's hoisted selection/onSelectionChange and pendingNav/onPendingNavResolved while adding lane-c's listState wiring and compactLabel/compactIcon.
- graph/StageDetailPane.kt and theme/Dimens.kt auto-merged and retain both sides' changes.
- Incoming branch af68a62 preserved as second parent.
Consolidate the omni-search row type badge and the stage-type chip into a
single ui/components/LabelChip composable, so the chip's shape, colour and
type live in one place and a second copy cannot drift.
Remove the unused SearchTarget.Run.artifacts field: an artifact hit and a
step hit now resolve identically to the visit's stage pane, whose fixed
section order already leads with the artifacts. Correct the
SearchUiState.failed KDoc to describe how prior results are kept, and
update OmniSearchResultsTest to the simplified target.
The pipelines list carried its own "Search pipelines" field while the app also
has the omni search, so the list offered a second, narrower way to do the same
job. Remove that field and move name narrowing into the list's filter control,
keeping "filters narrow one list" separate from "search finds across the system".
- PipelineListScreen no longer owns a `query` state or draws the
OutlinedTextField; the list is now the first child under the app bar, so the
reclaimed space comes from layout with no compensating spacer or new dp value.
- PipelineFilters gains a free-text `name` facet beside presence/lastRunStatuses,
applied by filterPipelines through the new pure pipelineMatchesName
(case-insensitive contains, blank means any). It needs no run history, so it
applies as soon as pipelines load while the run facets still wait for runsKnown;
activeCount and isActive include it, and hasRunFacets gates just the run facets.
- PipelineFilterControl opens with a Name section holding a single-line
PipelineNameFilter labelled "Pipeline name contains"; the shared Clear filters
resets it with the run facets and the active-count badge counts it.
- SettingsRepository/AppViewModel persist the name under pipeline_filter_name.
- Empty-state messages reflect the name-only miss.
- ListFilterTest covers the new name narrowing (+4 cases) and storage parsing.
- android/README.md documents the change and the build/APK evidence.
The runs list never had an inline search field, so nothing was removed there.
Add the fourth Search destination to the floating bottom container and
morph that container into the omni-search field.
- PipelinesApp: the destination pill gains a Search icon; tapping it
animates the icons out and turns the same Surface into a focused text
field whose size and position derive from the pill's own layout and the
window insets rather than fixed pill sizes. Back, the back gesture and a
scrim tap play the reverse and restore the pill with its previous
destination still selected. The open field covers the content behind a
scrim and shows a results panel above it.
- OmniSearch: pure state model (open/focused/query) with the open, dismiss,
focus and query transitions and a Bundle-able Saver, so an open field
survives a saved-state round trip.
- OmniSearchResults / SearchResultsPanel: map the typed search reply to
rows and render the ranked results, keeping the last accepted results
while a newer query loads and reporting a failure without blanking.
- AppViewModel / PipelinesRepository: drive queries over the existing
transport with debounce and reply ordering; the server search command
arrives in a later item, so a reply is classified by envelope type and a
non-search_results reply is reported with a clear message instead of
being mis-parsed, never a crash.
- Message: SEARCH / SEARCH_RESULTS constants and the SearchRequest,
SearchResponse and per-kind result payloads.
- AndroidManifest: adjustNothing so the app reads the IME from Compose
window insets and the field is not cleared twice by a window resize.
- Dimens: spacing for the search surface.
- Unit tests for the state model (pill, opening, focused, dismissed and
restored after the Saver round trip) and for the result-row mapping.
Mirror the server's derived verdict onto RunStep as a nullable
StageVerdict ({option, reason, reason_source, artifacts}, absent -> null),
matching internal/protocol/verdict.go.
Render a StageVerdictBlock at the top of the run pane, below the visit
tabs and above the facts, only when a verdict is present - a null or
empty verdict draws nothing, never an empty heading. The chosen option is
a ColoredChip in the stage outcome's existing colour; the reason is plain
pipelinesText().body that wraps without a maxLines clip.
The artifacts the reason names become inline, tappable references
(image/video open the shared ArtifactViewer, anything else uses the
existing openArtifact action) plus the same references as compact
FlowRow chips. Referenced names resolve to the visit's recorded
artifacts, deduplicated and in recorded order; reason text is split
around those names longest-match-first. No new dependency and no new dp
or sp literal.
android/ only.
The covered panes spent a fixed 224dp (PEEK_BUDGET_DP) whatever the window
width, so a folded window lost over half its width and the active pane was
squeezed as levels were added. Derive the covered region at runtime from the
compact identity a covered edge must show - an icon plus the shortest label
that identifies the pane, measured with the app's own type role and floored at
the platform minimum touch target - plus the leading gesture inset.
coveredEdgesDp now returns edges that sum to exactly that region, so the active
pane opens at hostWidth - region and its width is identical at every depth:
opening a level only adds a thin edge inside the region. The nearest covered
level shows its identity, one older level is a thinner touch-target edge, and
older levels fold into a count rather than accumulating width; tapping an edge
still pops everything above it. Removes PEEK_BUDGET_DP, MIN_ACTIVE_PANE_WIDTH_DP
and the peekBudgetDp/peekStripsDp budget split from Dimens.kt and
SlidingPaneStack.kt. PaneIdentity gains the pane-kind icon drawn above the title
at the vector's own measured default size, and each destination passes its
compact label/icon.
The touched files also carry the in-flight covered-pane statistics and list-scroll
plumbing they depend on (PaneStat/paneStatTier/visiblePaneStats, per-destination
LazyListState); the remaining pieces of that work remain uncommitted.
Exchange a build/protocol identity so an older CLI can no longer talk to a
newer native server silently:
- protocol carries ProtocolVersion and every server frame (including the
unsolicited welcome) is stamped with the server build and protocol, so a
client that never pings still learns the server identity.
- buildinfo.Compare classifies the relationship as InSync/Warn/Refuse and
names both builds; the client Watch reports one warning per connection and
returns a hard error when the server speaks a newer, uninterpretable
protocol so the caller never acts on the frame.
- client.ReceiveNotice decodes the raw frame so an unmodeled top-level field
produces a one-time "unrecognized field(s)" notice instead of being dropped
silently (the case that hid the artifacts field).
- CLI routes warnings to stderr, keeps machine JSON stdout clean and fails
non-zero on a refused mismatch; ping/connection print the client/server
relationship. The TUI surfaces the same notices through the shared log.
- scripts/setup-native.py now warns when an older `pipelines` earlier on PATH
shadows prefix/bin/pipelines, naming both paths and versions, without
installing or modifying anything else.
- Docs, the packaged debugging skill reference and README no longer imply the
installed CLI always matches the server.
Tests cover the comparison table, unknown-field detection and caching, the
server identity stamping on pong and welcome, the CLI/TUI reporting paths and
the installer PATH shadowing warning.
Surface the operator's title for a run so two runs of the same pipeline
read as clearly different goals, with the pipeline name kept as context.
- protocol: optional nullable `title` on RunSummary and RunDetail; absent
or blank degrades to the pipeline name.
- ui/RunIdentity.kt: one shared fallback (`runLabel`) and two-line identity
(`runIdentity`) so the list, detail header and stage pane never spell a
run differently; a titled run shows the pipeline name + version as a
second line, an untitled run keeps the pipeline name as its only line.
- runs list / run detail / stage pane use the shared identity; the status
chip, duration and metadata layout are unchanged.
- no rename affordance: the server exposes no title and no `set_run_title`,
so the app does not fake renaming.
- tests: RunIdentityTest (5) plus ProtocolModelTest cases for a present and
an absent title; README documents the change and its verification.
A running run now shows how long it has taken so far instead of its start
time, kept live: the runs-list meta row, the pipeline list's last-run line
and the run detail summary's Duration row all read one shared rule,
runDurationText (ui/Format.kt), which reports the elapsed while executing and
keeps the stored total once the run has finished.
The elapsed is derived from the server's own start timestamp against a single
shared clock, rememberLiveNowMillis/hostResumed (ui/components/LiveTime.kt),
so it stays correct after a reconnect and never ticks while the app is
backgrounded or while no shown run is executing. WorkflowGraph's private
lifecycle read and once-a-second loop are replaced by that same shared pair.
A missing start omits the value and a future start (clock skew) reads as 0s.
Unit tests (FormatTest, PipelineLastRunTest) cover a run that has just
started, the minute and hour boundaries, a missing and a future start, and a
finished run keeping its stored total duration.
The run summary led with the artifacts gallery, but it was always open, so a
long run's grid pushed the run's own facts below the fold. The gallery now
sits in one collapsed-by-default section in the same place, with a header that
describes itself while collapsed.
- RunDetailScreen.kt: RunSummaryTab renders the gallery through the shared
LazyListScope.collapsibleSection, keyed "artifacts" and seeded with
rememberSectionExpansion(run.id), so a run opened fresh starts collapsed and
the operator's choice persists per run across tab switches and config
changes. Order is unchanged: the section still leads the summary and the
facts item follows it. Expanding composes the unchanged ArtifactGallery;
collapsing composes nothing.
- ArtifactGroups.kt: new pure runArtifactsHeader(steps, executing) builds the
self-describing header - singular/plural title, recorded count and
"· partial capture" when any visit excluded or omitted files; empty runs
reuse the shared artifactsEmptyMessage. Its count equals runArtifactGroups'
tile total, so the header and the expanded gallery cannot disagree.
- ArtifactGroupsTest.kt: four header tests covering count, singular vs plural,
the partial-capture notes and the empty case, plus a count-matches-gallery
assertion.
- android/README.md: records the change and its verification evidence.
Android only; no new dependency, no run started from the app, device work on
the operator's Pixel 10 Pro Fold (58281FDCG000JF). Verified: 269/0/0/0 unit
tests, clean assembleDebug and published APK LAN URL 200.
Refinements on top of the human decision stage (a6aff47):
- `pipelines decide` now reports the feedback the server actually stored for
the chosen option, so a repeated (idempotent) decision shows the recorded
evidence instead of the ignored request text. A repeated decision is covered
by a new assertion that also checks the step output does not change.
- The decision option/feedback output keys are defined once in protocol
(DecisionOptionKey/DecisionFeedbackKey) with a DecisionEvidence reader, and
the store and CLI use them instead of bare string literals.
- piEnvironment drops any inherited PIPELINES_ARTIFACTS_DIR before spawning Pi,
so the artifactsDir argument is authoritative and an ambient directory (for
example Pipelines' own when a stage runs inside another stage) never leaks
into a child that was given none. Covered by a test with an ambient value set.
Standardise the artifacts gallery so the run summary and the stage pane
render one tile per artifact instead of images as a grid and everything
else as a plain list.
- Artifacts.kt: ArtifactTile now covers both shapes - an image/video fills
the cell with a thumbnail (a video shows its poster frame with a play
badge), anything else shows its file-type icon with the name and size
beneath it; the grid, share/copy actions and multi-select are unchanged
and the capture wording is untouched.
- ArtifactType.kt (new): pure ArtifactKind classification from content
type plus name extension fallback, and artifactIcon(kind) mapping each
kind to a distinct glyph.
- ArtifactImage.kt: ArtifactVideoFrame/rememberArtifactVideoFrame extract
a poster frame via MediaMetadataRetriever over an in-memory
MediaDataSource; the shared memoized-preview scaffold is factored into
rememberArtifactPreview.
- ArtifactViewer.kt: full-screen viewer takes a video flag and shows the
poster frame.
- Format.kt: add isVideoContentType; drop the now-unused artifactMetaLine.
- Dimens.kt: ARTIFACT_TYPE_ICON_DP, ARTIFACT_PLAY_BADGE_DP,
ARTIFACT_TILE_ACTION_DP tokens.
- Tests: new ArtifactTypeTest (type-to-icon mapping, extension fallback,
distinct glyphs); FormatTest updated for the removed helper.
- android/README.md documents the pass.
android/ only; no new dependencies beyond material-icons-extended.
Expanding a stage section rendered its rows stacked at the top of the section: CollapsibleSectionBody passed the section's several field composables straight into AnimatedVisibility, which measures a single child, so all rows were placed at the section origin and only the last (Memory) was visible. Fix the shared body by laying its content out in one full-width Column, and make every stage section open as its own pane rather than growing the stage screen inline.
- PaneState: add the Section pane level (MAX_PANE_DEPTH = 3), a saved sectionKey primitive, and the matching backStep/popTo/encode/decode handling so one Back or peek pops one level and the open section survives rotation and process death.
- StageDetailPane: the stage screen now keeps a compact collapsed list of sections (Artifacts, Output, Input, This run, Stage definition); tapping one opens it as a full-window screen on a narrow window or a panel to the right of the stage pane on a wide one, with its own title, scroll and top spacing. A shared stageSectionSpecs derives the sections for both the stage screen and the section pane, and the artifacts section keeps the shared ArtifactSelectionScaffold.
- PipelinesApp: wire the Section pane into the stack, its peeks and Back. Section content (Definition, Input, Output, Artifacts) and semantics are unchanged; sections start collapsed and the operator's choice persists.
- Tests: cover the section pane level, backStep, popTo and the shared stageSectionSpecs. Build and forced unit tests pass (262 tests, 0 failures); lint shows only the pre-existing themes.xml NewApi errors.
Android only (7 files); no dependency added, no server/Go/installer/Docker change, no run started from the app. APK published to the run artifacts directory and documented in android/README.md.
Extend ChatApprovalTest so once, always and reject each go through the same
ChatProtocol.replyRequest builder plus one encode/decode round trip using the
DECISION_* constants, so no decision is asserted more weakly than another, and
add answeringAlwaysUpdatesTheCardInPlace (settled card, no actions, risk panel
retired, request still visible). Refresh the android/README.md verification
notes: 267 tests, the rebuilt APK's sha256 and byte size, the 147/3 artifact
path and LAN URL, and the emulator/device-lock notes for this pass.
A new registered stage type `human` (internal/stage/human.go) has no model,
agent, tools or configs: its required `prompt` block carries the operator
instructions and its non-error edges are the selectable options. It parses,
validates, renders, exports and imports like every other type (no parser
special-case, no config keys).
At runtime the runner parks the run instead of executing: the step and the run
persist `waiting` with self-contained {prompt, options} evidence, and the run
returns without starting later stages, capturing file diffs, scanning artifacts,
holding a goroutine/timer or applying a stage timeout. The wait is durable: a
restart leaves the waiting rows untouched (RecoverRuns excludes them) and a
rebuilt runtime over the same database still accepts the decision.
`pipelines decide <run-id> [stage] --option KEY [--feedback TEXT] [--json]` and
the TUI `c` picker resolve the decision from the pinned graph in one store
transaction, record the choice and feedback as the step's evidence and resume
along edges[KEY]. An unknown option is refused with the valid list, a stage that
is not waiting is refused, and a repeated decision for an already-decided visit
does not double-apply. Cancel and retry keep working while waiting and after a
decision; a human stage is never killed by a stage timeout or max_visits, and a
`notify` policy can fire an `on=waiting` event.
Docs: the authoring skill references (format.md, operations.md), docs/cli.md and
README describe the stage, the command and the TUI picker. Tests cover stage
validation, the parsing round-trip, the store transitions, a websocket wait/decide
integration, cancel/retry, waiting notifications and a persistence-layer rebuild
(second Open + newRunner over one database, no process restart).
While a turn was in flight the trailing composer button became the stop
control, but the text field's keyboard send action still checked only
canSend, so a prompt could be sent from the IME during a run even though
the visible control had become stop.
Add ChatFormat.composerAction(thread) -> ComposerAction.Send/Stop as the
single pure derivation and read it from both the trailing button and the
keyboard action, so the two controls can no longer disagree.
ChatFormatTest.composerControlIsStopWhileARunIsInFlightAndSendOtherwise
covers Send when idle and Stop while running. android/ only; no new
dependency.
Render a server-pushed chat_permission as an approval card in the
transcript instead of a transient bar above the composer, matching the
reference opencode/Jarvis chat: the requested action, its resources, the
amber RISK REVIEW panel, and once/always/reject actions wired to the
server's chat_reply.
- ChatFormat: a pure approvalCard()/ApprovalCard model and one
ChatProtocol.replyRequest() builder shared by the card's actions and
the repository, so the wire frame and the controls cannot drift.
- ChatStore/AppViewModel: answerPermission() settles the card in place
optimistically and survives a chat switch; a failed reply reopens the
same request rather than showing an unacknowledged decision.
- ChatPermissionFrame gains optional resolved/decision fields so a
request the server reports as already settled shows its verdict with
no stale actions; a new request replaces a settled card.
- Findability: a gold PanTool top-bar action and chat-chip marker jump
to the pending card; the running indicator stands down while waiting.
- Tests: ChatApprovalTest covers pending/answered/resolved, the reply
builder's once/reject decisions and the missing-summary fallback;
ChatProtocolTest covers the server-resolution frame. README records
the 265-test/assembleDebug evidence and the emulator-5556 captures,
and .approval_check/chat_stub_approval.py is the test double used.
android/ only; no new dependency beyond material-icons-extended; no
Pipelines server, Go, installer or Docker file touched; no run started
from the app. The live brokered server is not verified (the resolved
frame fields are an additive contract it does not emit yet).
One query now finds anything in the system over the wire and from the CLI.
- protocol: add Search/SearchResults message types and a SearchRequest
(query, optional kind filter, optional limit) / SearchResponse (typed,
ranked matches plus truncation and scan-bound counters) pair, with typed
payloads for pipeline, version, run, step/visit, field text and artifact.
- store: add Store.Search, a read-only query over pipelines,
pipeline_versions, pipeline_runs, run_steps (+step_control) and
step_artifacts. Ranks exact name (3) > name substring (2) > body text (1),
breaks ties by recency then a stable kind/target order, and reports the
matched field. Bounds work with a 50-default/200-max result cap, a 2000-row
per-kind scan limit and a 2000-step / 1 MiB field-text scan budget.
- terms: a bare status word or lone date expression (today, yesterday, this
week, YYYY-MM-DD) selects runs; anything else stays free text.
- server: dispatch protocol.Search to stores[0].Search; read-only and
unconfirmed.
- cli: add `pipelines search QUERY [--type KIND] [--limit N] [--json]`, help
text and a BROWSE line.
- store: add OpenReadOnly so the live-database check never runs DDL.
- docs: document the grammar, kinds, ranking and bounds in docs/cli.md,
scope the TUI omnisearch to pipelines in docs/ui-ux-review.md, and add a
search bullet to the packaged operations reference.
- tests: protocol round trip, store/ranking/term-parsing/limit/scan-bound
unit tests, a server end-to-end test, and an env-guarded live check.
No new dependencies; go.mod/go.sum unchanged.
Extract the single chat_interrupt frame into ChatProtocol.interruptRequest and
have PipelinesRepository.chatInterrupt send it, so production and its unit test
share the one wire shape instead of the repository building it inline.
Add chatInterruptRequestCarriesTheRunningSession, which asserts the frame the
stop control sends is chat_interrupt carrying the running session's id (and no
chat prompt/reply) and that it survives the client's JSON round-trip.
Document the slice and its emulator-5556 evidence in android/README.md.
The chat edit diff card rendered every line of an edit, so the >400-line
fallback flooded the transcript with thousands of Text nodes on a phone;
it also showed only the tool name, not the file, on the expanded panel.
- ChatFormat.toolEditDiff keeps the first MAX_RENDERED_DIFF_LINES (200)
lines and puts the remainder in the new ToolDiff.omitted, while the
+N/-M counts continue to describe the whole edit.
- DiffView prints "… N more changed lines not shown" first so the
truncation is visible without scrolling.
- The expanded panel header now shows the edited file name and the shared
green/red DiffStat, matching the collapsed chip.
- ChatFormatTest pins omitted = 0 for a small edit and the true counts,
200-line cap and omitted = 300 for an oversized one.
- android/README.md records the change and its evidence.
Move a completed stage visit's derived verdict out of the Progress summary
and into the top of the selected stage's Result section, so the chosen
option, its concise reason and the referenced artifacts are visible where
the stage is inspected.
- Add stageVerdictLines, which renders one line "Verdict: <option> · <reason>"
(still stating the option when the reason is absent) plus a following
"Referenced artifacts:" line only when artifacts exist, and returns nil
for a nil verdict so nothing extra is shown.
- Prepend those lines in runSectionLines after the stage state and before
the stage output and evidence, without adding a section or changing the
panel layout.
- Update the TUI test to render through the run-detail view, asserting the
option, reason and artifact names appear, and that a nil verdict adds
nothing; document the TUI placement in the debugging skill reference.
Omni-search work already present in the working tree (including the
runDuration→protocol.Duration refactor in run_detail.go) is unrelated to
this change and is left uncommitted.
Render each assistant turn's tool calls the way the reference jarvis chat
does: one compact chip per call showing the tool name, its duration and a
spinner while it runs, with consecutive calls of a turn grouped into one
wrapping row so the transcript stays scannable (ChatScreen AssistantParts /
ToolCalls).
Tapping a chip opens a floating panel over the transcript with INPUT as
pretty-printed JSON and OUTPUT as monospace, plus copy actions. The panel is
dismissible by tapping outside or its Close action and is positioned by
ToolPanelPositionProvider so it never moves the conversation (the transcript
scroll position is untouched while the overlay is open).
A failed call is shown as failed and renders both its diff (when present) and
an ERROR block. ERROR text now comes from the new toolErrorText, which prefers
the server's error message over any partial output, while toolOutputText is the
OUTPUT derivation again; toolFailed identifies the error state.
While a turn streams, the transcript follows the newest content with a hard
jump unless the operator has dragged away, driven by
listState.interactionSource.interactions (DragInteraction.Start/Stop/Cancel)
rather than programmatic scrolls. The running indicator and its live elapsed
time derive from the turn's own start (ChatThread.elapsedMillis, an injectable
clock in ChatStore) so they survive a chat switch.
A reconnecting socket now settles every open chat and re-fetches its history
(ChatStore.streamResumed, AppViewModel.observeConnection), and a server-pushed
stream error with no session id settles the sole/active running turn with the
server's message instead of leaving the indicator spinning.
Tests: ChatFormatTest covers toolErrorText; new ChatStreamTest covers the
streaming/elapsed/reconnect/stream-error guarantees against an injected clock;
new ToolPanelPositionProviderTest covers the panel's below/flip/margin
placement. README's chat section documents the follow-the-stream mechanism.
Each completed step now exposes a verdict: the chosen option (outcome -> target),
a bounded one-line reason with model/fallback/absent provenance, and the recorded
artifacts the reason refers to. The store persists only the model-generated line
in a new step_verdicts table and derives the fallback at read time (no backfill);
the server generates the line once at completion, best-effort; the TUI renders it
in the stage progress summary. The packaged debugging reference documents it.
Add a bounded, never-empty title to every run. `pipelines run --title "..."`
sets it explicitly and skips model generation; otherwise the server generates
it from the prompt through the shared summariser and falls back to the prompt's
first line, recording title_source (model/fallback/explicit) and title_model.
- protocol/store: RunSummary.Title plus Label(), a set_run_title message, title
columns with an idempotent migration, CreateRunTitled, and SetRunTitle renames.
- summarise: Title/GenerateTitle/DeriveTitle/ExplicitTitle bounded to TitleCap.
- runner: generate the title outside the admission lock and only for a request
that starts a run; a replay reuses the stored title and an explicit title
skips the model entirely.
- CLI: `--title` on run and a `title <run-id> "<title>"` rename command; human
`runs` and `inspect` show the title beside the pipeline name and version.
- TUI: the run list row, run detail header, Home run cards and last-run line,
and the pipeline Results line show the title via Label().
Tests cover storage/migration/fallback, explicit-title and rename, the
end-to-end CLI flow, and title rendering; CLI and operations docs updated.
Move the workflow-graph package (WorkflowGraph, the StageDetailPane sheet,
Artifacts, ArtifactViewer, ArtifactImage) off its remaining raw dp literals
and onto the geometry tokens, all substitutions value-equal so the visuals
are unchanged.
- ui/theme/Dimens.kt gains the off-scale component and graph-chrome sizes:
the run legend's GRAPH_LEGEND_PAD_X/PAD_Y/GAP, the collapsed
GRAPH_PANEL_MIN_HEIGHT (moved off WorkflowGraph.kt, now imported by its
test), the gallery's ARTIFACT_TILE_DP/THUMB_DP/BADGE_DP/PROGRESS_DP/
PROGRESS_STROKE_DP, and the stage-type chip's STAGE_CHIP_PAD_H.
- Artifact tile/row and thumbnail corner radii derive from the Material
shape scale (MaterialTheme.shapes.medium/.small), matching the graph's
existing node corners.
- DimensTest.kt's allow-list shrinks from 24 to 21 entries: 24.dp, 56.dp and
96.dp are gone now that their last call sites migrated. The structural
hairlines/zeros, the not-yet-migrated app shell and Settings literals, and
the sp type values owned by Typography.kt remain.
- WorkflowGraphTest.kt imports the moved GRAPH_PANEL_MIN_HEIGHT.
- android/README.md documents the change and its verification evidence.
Verified: gradle :app:assembleDebug :app:testDebugUnitTest --rerun-tasks
BUILD SUCCESSFUL, 30 XMLs / 259 tests, 0 failures/errors/skipped; the
enforcement test fails on a deliberately introduced literal and passes once
removed; before/after captures at both window widths differ only on the
status-bar clock and live relative-time text.
A model stage whose final result was malformed failed the whole run even when
the work and evidence had survived, because the only recovery was an isolated
formatter call on a fixed 30-second deadline that could time out on its own
(run 55, ui_reviewer of pipeline 9 v6).
Add an ordered, deterministic local repair pass (internal/stage/result_repair.go)
that runs before any model call: (a) append missing closing delimiters via the
existing closeJSONContainers; (b) remove one unambiguous trailing comma before a
closing } or ]; (c) insert exactly the single comma/colon the JSON parser reported
at its byte offset. Every candidate must pass losslessSyntaxRepair(original,
candidate) and decodeModelResult before it is accepted, so no field value is
invented, removed, reordered or rewritten. Ambiguous payloads (an unescaped quote
or colon inside a string value, or a stray delimiter with no single completion)
return not-ok and fall through to the bounded model re-emit rather than guessing.
stageResultText still strips surrounding prose and a single code fence first;
model_prompt.go now also strips prose around a single balanced-but-malformed
object span so the deterministic pass sees just the object.
Replace the fixed 30-second formatter deadline with repairDeadline(payloadBytes,
remaining), which grows monotonically with payload size, has a floor, and is
clamped to the attempt's remaining budget minus a safety margin, so the stage
context remains the outer bound and a large result is not guaranteed to time out.
The one-request-per-attempt, 32 KiB bound and the deny-all, retained, new-session
property of RepairFormat are unchanged for OpenCodeClient and PiClient.
Record an explicit recovery status on every terminal path: repaired, partial or
unrecoverable, plus a byte offset. A candidate that is valid JSON but still fails
envelope validation is marked partial and never returned as a completed result;
the unrecoverable path retains the bounded original payload, the precise parse
error (including its byte offset) and surfaces the status in the FINAL RESULT
VALIDATION diagnostic.
safeToReplay is intentionally unchanged. The deterministic local repair and the
fresh deny-all, no-tools, retained formatter session are independent of the
stage's tool set and never re-prompt or replay the admitted agent session, so a
shell-enabled review stage gains malformed-result recovery without any replay of
admitted work. Risk: no change to mutation safety; the only added cost is one
bounded formatter call, already budgeted. Replayability is not extended for shell
stages.
Docs and packaged skill references (docs/model-results.md,
internal/author/skill/references/{format,debugging,verification}.md) describe the
three-tier recovery, the status words and the derived deadline; the safeToReplay
decision and its risk are recorded there alongside the commit.
Tests add mocked, offline coverage for the deterministic and partial paths, a
repairDeadline monotonicity/floor/clamp table, and a shell-stage regression
proving a recoverable malformed result does not fail the run or replay tool work.
Add a dependency-free CommonMark-subset parser/renderer (ui/chat/Markdown.kt):
paragraphs, headings, fenced code, block quotes, bullet/ordered lists, tables,
thematic rules and inline code/bold/italic/links/autolinks. The renderer turns
blocks into themed, selectable Compose content: fenced code reuses the shared
CodeBlock (existing whole-block copy action), tables are a bordered grid with a
bold header, links use LinkAnnotation.Url, and inline code/headings use the
shared type roles.
ChatScreen.MessagePart now calls MarkdownText(part.text) for PART_TEXT instead
of a single plain Text; the assistant message stays one SelectionContainer, so
all of it remains selectable and long content stays inside the bubble. Geometry
reuses the chat tokens (CHAT_ITEM_GAP promoted private -> internal).
Add MarkdownTest with 10 unit tests covering fences/language, unterminated
fences, soft-wrap joining, headings/rules, lists, tables, inline spans, bare-URL
autolinks and snake_case non-emphasis. Document the change in android/README.md.