diff options
| author | soryu <soryu@soryu.co> | 2026-05-16 19:55:34 +0100 |
|---|---|---|
| committer | soryu <soryu@soryu.co> | 2026-05-16 19:55:34 +0100 |
| commit | 8cd7b40ace4e5e2b22ad89aafec74c7655def19b (patch) | |
| tree | 73e9ba4b91086cf8043eb71a295f75589e9bbe90 /makima/frontend/src/components/directives/DocumentTaskStream.tsx | |
| parent | 8e2bbcab1a7b3b9005803d7ce3bfce7fa483a4d7 (diff) | |
| download | soryu-strict-orchestration.tar.gz soryu-strict-orchestration.zip | |
feat(directives): strict orchestration flow + sidebar overhaul + task page rewritestrict-orchestration
End-to-end rewrite addressing the issues from the user's UX review.
The system now feels like a daemon-orchestration tool: lock a contract
and the orchestrator just goes; PR raised → auto-ship → reopen for
amendments. The sidebar tree shows real entities only (no duplicates,
no inline action buttons polluting the file list), and every entity
gets a right-click context menu. Task page matches the old /exec
layout (diff on the left, feed + composer on the right).
## Backend — strict lifecycle (the orchestrator-never-spawned bug)
Root cause: `phase_planning()` gates on `directive.status='active'`, but
`start_contract()` only flipped the contract row — the parent directive
stayed in whatever state it was. So locking a contract did nothing
visible.
Fix: contract lifecycle now drives directive status in the same
transaction.
start_contract → if contract becomes active, flip directive
draft|paused|idle|inactive → active
pause_contract → after promote, if no active contract left,
directive → paused
complete_contract→ after promote, if no active left, directive →
inactive (also fires on auto-ship from PR detect)
unlock_contract → if was active and no active left, directive →
paused
reopen_contract → NEW. shipped → active. Directive → active,
orchestrator_task_id/pr_url/pr_branch cleared so
the reconciler spawns a fresh planner. The
planner reads get_latest_merged_revision and
frames the new plan as an amendment.
handlers::directive_documents lifts state.kick_directive_reconciler()
into run_contract_transition so every successful transition wakes the
reconciler immediately (no 15s wait).
handlers::directives `update_directive` (PR-detection branch) calls
`complete_contract(active_contract_id, pr_url, pr_branch)` instead of
`set_directive_inactive`. The contract auto-ships; the directive
follows via the sync above. No more manual "Mark complete" click.
POST /api/v1/contracts/{id}/reopen added + wired through openapi.
Spawn task names dropped the directive-title prefix that looked
redundant in the sidebar:
"Plan: <title>" → "orchestrator"
"Re-plan: <title>" → "orchestrator (re-plan)"
"PR: <title>" → "completion"
"Update PR: <title>" → "completion (update)"
## Frontend — sidebar
* De-dupe: DocumentTasksFolder filters tasks[] to exclude any task
whose id already appears in steps[].taskId. Single row per task,
single highlight on click.
* Generic SidebarContextMenu (new) replaces the directive-only
DirectiveContextMenu (deleted). Per-entity item arrays built at the
page level — directive, contract, step, task each have their own
contextual actions.
* Right-click works on every sidebar entity now (was directive-only).
* `+ New document` / `+ New ephemeral task` inline buttons removed.
Reachable via the directive folder right-click OR the hover-only
`+` button on the directive folder row.
* ContractHeader: dropped "Mark complete" button (auto-fires on PR).
Added "Reopen for amendment" button when contract is shipped.
## Frontend — task page rewrite
TaskPage.tsx replaces DocumentTaskStream.tsx (deleted). Two-column
layout matches the old /exec page that the user preferred:
┌────────────────────────┬──────────────────────────────────┐
│ Changed files (~30%) │ Transcript feed (scrollable) │
│ ────────────────── │ ────────────────────── │
│ src/foo.rs │ [user] do thing │
│ src/bar.rs │ [tool] Read foo.rs │
│ │ │
│ Diff (selected file) │ │
│ ├──────────────────────────────────┤
│ │ Composer (sticky bottom) │
└────────────────────────┴──────────────────────────────────┘
Diff comes from getTaskDiff(); parseDiff + DiffFileView exported from
OverlayDiffViewer for reuse (no duplication). Diff auto-refreshes
when the task transitions to a terminal state. Transcript styling +
sticky composer keep the parts the user liked. "Open in task page"
button removed — the right pane IS the task page.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Diffstat (limited to 'makima/frontend/src/components/directives/DocumentTaskStream.tsx')
| -rw-r--r-- | makima/frontend/src/components/directives/DocumentTaskStream.tsx | 516 |
1 files changed, 0 insertions, 516 deletions
diff --git a/makima/frontend/src/components/directives/DocumentTaskStream.tsx b/makima/frontend/src/components/directives/DocumentTaskStream.tsx deleted file mode 100644 index b718ae4..0000000 --- a/makima/frontend/src/components/directives/DocumentTaskStream.tsx +++ /dev/null @@ -1,516 +0,0 @@ -/** - * DocumentTaskStream — renders a running task's output as a flowing document - * (assistant prose, tool blocks) instead of the boxy log style of TaskOutput. - * - * Key differences from TaskOutput: - * - Document typography (serif-ish paragraphs, not monospace logs). - * - Interleaved with subtle marginalia for tool calls and results. - * - Sticky comment composer at the bottom that's always in view. - * - Header strip with explicit Stop / Send / Open-in-task-page buttons so - * primary task controls don't require a right-click discovery step. - * - Module-level cache of historical entries per taskId so re-selecting a - * task you've already viewed renders instantly while a fresh fetch - * refreshes in the background. - */ -import { useCallback, useEffect, useRef, useState } from "react"; -import { useNavigate } from "react-router"; -import { SimpleMarkdown } from "../SimpleMarkdown"; -import { - useTaskSubscription, - type TaskOutputEvent, -} from "../../hooks/useTaskSubscription"; -import { getTaskOutput, sendTaskMessage, stopTask } from "../../lib/api"; - -interface DocumentTaskStreamProps { - taskId: string; - /** Human label used as the document header (e.g. "orchestrator", step name) */ - label: string; - /** - * When this task is ephemeral (spawned via the directive's "+ New task" - * action) AND has reached a terminal state, surface a "Merge to base" - * affordance that navigates the user to the standalone task page where - * the existing merge UI handles the actual merge / conflict flow. - * - * Step-spawned tasks have their own merge path (the directive's PR), so - * this affordance is intentionally off by default. - */ - ephemeral?: boolean; - /** Current status of the task; drives whether merge button is enabled. */ - status?: string; -} - -// ============================================================================= -// Module-level cache for historical task entries. -// -// Switching between tasks you've already viewed used to re-fire -// getTaskOutput and show "Loading transcript…" for the duration of the -// network round-trip. We now keep the entries cached per taskId; on -// re-selection we render the cache immediately and refetch in the -// background. The WS subscription continues to handle live deltas. -// ============================================================================= -const entriesCache = new Map<string, TaskOutputEvent[]>(); - -export function DocumentTaskStream({ - taskId, - label, - ephemeral, - status, -}: DocumentTaskStreamProps) { - const navigate = useNavigate(); - const [entries, setEntries] = useState<TaskOutputEvent[]>( - () => entriesCache.get(taskId) ?? [], - ); - const [loading, setLoading] = useState(!entriesCache.has(taskId)); - const [isStreaming, setIsStreaming] = useState(false); - const [comment, setComment] = useState(""); - const [sending, setSending] = useState(false); - const [sendError, setSendError] = useState<string | null>(null); - const [stopping, setStopping] = useState(false); - const containerRef = useRef<HTMLDivElement>(null); - const composerRef = useRef<HTMLDivElement>(null); - // autoScroll lives in a ref so the scroll handler reads the latest value - // synchronously without re-creating the effect. - const autoScrollRef = useRef(true); - const [showResumeScroll, setShowResumeScroll] = useState(false); - - // Load historical output when the selected task changes. Render the cache - // immediately if we have it; refetch in the background regardless. - useEffect(() => { - let cancelled = false; - const cached = entriesCache.get(taskId); - if (cached) { - setEntries(cached); - setLoading(false); - } else { - setEntries([]); - setLoading(true); - } - setIsStreaming(false); - - getTaskOutput(taskId) - .then((res) => { - if (cancelled) return; - const mapped: TaskOutputEvent[] = res.entries.map((e) => ({ - taskId: e.taskId, - messageType: e.messageType, - content: e.content, - toolName: e.toolName, - toolInput: e.toolInput, - isError: e.isError, - costUsd: e.costUsd, - durationMs: e.durationMs, - isPartial: false, - })); - entriesCache.set(taskId, mapped); - setEntries(mapped); - }) - .catch((err) => { - if (cancelled) return; - // eslint-disable-next-line no-console - console.error("Failed to load task output history:", err); - }) - .finally(() => { - if (!cancelled) setLoading(false); - }); - return () => { - cancelled = true; - }; - }, [taskId]); - - const handleOutput = useCallback( - (event: TaskOutputEvent) => { - if (event.isPartial) return; - setEntries((prev) => { - const next = [...prev, event]; - entriesCache.set(taskId, next); - return next; - }); - setIsStreaming(true); - }, - [taskId], - ); - - const handleUpdate = useCallback((event: { status: string }) => { - if ( - event.status === "completed" || - event.status === "failed" || - event.status === "cancelled" || - event.status === "interrupted" || - event.status === "merged" || - event.status === "done" - ) { - setIsStreaming(false); - } else if (event.status === "running") { - setIsStreaming(true); - } - }, []); - - useTaskSubscription({ - taskId, - subscribeOutput: true, - onOutput: handleOutput, - onUpdate: handleUpdate, - }); - - // Auto-scroll while at bottom. The previous version only flipped autoScroll - // off and never resumed; now a scroll back into the bottom 80px reactivates - // it so a brief read-up doesn't permanently freeze the stream at the top. - useEffect(() => { - if (autoScrollRef.current && containerRef.current) { - containerRef.current.scrollTop = containerRef.current.scrollHeight; - } - }, [entries]); - - // After loading the initial transcript, snap to the bottom unconditionally - // so users see the latest output, not the start. - useEffect(() => { - if (!loading && containerRef.current) { - containerRef.current.scrollTop = containerRef.current.scrollHeight; - autoScrollRef.current = true; - setShowResumeScroll(false); - } - }, [loading, taskId]); - - const handleScroll = useCallback(() => { - if (!containerRef.current) return; - const { scrollTop, scrollHeight, clientHeight } = containerRef.current; - const distanceFromBottom = scrollHeight - scrollTop - clientHeight; - const atBottom = distanceFromBottom < 80; - autoScrollRef.current = atBottom; - setShowResumeScroll(!atBottom); - }, []); - - const submitComment = useCallback( - async (e: React.FormEvent) => { - e.preventDefault(); - const trimmed = comment.trim(); - if (!trimmed || sending) return; - setSending(true); - setSendError(null); - // Show the comment immediately as a user-input entry. - setEntries((prev) => { - const next: TaskOutputEvent[] = [ - ...prev, - { - taskId, - messageType: "user_input", - content: trimmed, - isPartial: false, - }, - ]; - entriesCache.set(taskId, next); - return next; - }); - try { - await sendTaskMessage(taskId, trimmed); - setComment(""); - } catch (err) { - setSendError( - err instanceof Error ? err.message : "Failed to send comment", - ); - window.setTimeout(() => setSendError(null), 5000); - } finally { - setSending(false); - } - }, - [comment, sending, taskId], - ); - - const handleStop = useCallback(async () => { - if (stopping || !isStreaming) return; - if (!window.confirm("Stop this task? It will be marked failed.")) return; - setStopping(true); - try { - await stopTask(taskId); - } catch (err) { - // eslint-disable-next-line no-console - console.error("Failed to stop task", err); - } finally { - setStopping(false); - } - }, [taskId, stopping, isStreaming]); - - const focusComposer = useCallback(() => { - const input = composerRef.current?.querySelector("textarea"); - input?.focus(); - }, []); - - const resumeScroll = useCallback(() => { - if (!containerRef.current) return; - containerRef.current.scrollTop = containerRef.current.scrollHeight; - autoScrollRef.current = true; - setShowResumeScroll(false); - }, []); - - return ( - <div className="flex-1 flex flex-col h-full overflow-hidden bg-[#0a1628] relative"> - {/* Action header strip — explicit Stop / Send / Open-in-task-page so - users don't have to right-click to discover task controls. */} - <div className="shrink-0 flex items-center gap-2 px-6 py-2 border-b border-dashed border-[rgba(117,170,252,0.2)] bg-[#091428]"> - <span className="text-[10px] font-mono text-[#556677] uppercase tracking-wide"> - Task actions - </span> - <button - type="button" - onClick={focusComposer} - className="ml-auto px-2 py-1 font-mono text-[10px] uppercase tracking-wide text-emerald-300 border border-emerald-700/60 hover:border-emerald-400" - > - Send (⌘↵) - </button> - <button - type="button" - onClick={handleStop} - disabled={!isStreaming || stopping} - className="px-2 py-1 font-mono text-[10px] uppercase tracking-wide text-amber-300 border border-amber-600/60 hover:border-amber-400 disabled:opacity-40 disabled:cursor-not-allowed" - > - {stopping ? "Stopping…" : "Stop"} - </button> - - {/* Manual merge affordance — visible only on ephemeral tasks that - have reached a terminal state. Navigates to the standalone task - page where the existing mesh_merge UI drives the real merge / - conflict resolution flow. The user explicitly asked for this to - be a manual button press for safety. */} - {ephemeral && isTerminalStatus(status) && ( - <button - type="button" - onClick={() => { - const ok = window.confirm( - "Merge this ephemeral task into the base branch? You'll be taken to the task page where the merge runs and any conflicts are resolved.", - ); - if (!ok) return; - navigate(`/exec/${taskId}#merge`); - }} - title="Manual merge — opens the merge UI on the task page" - className="px-2 py-1 font-mono text-[10px] uppercase tracking-wide text-emerald-300 border border-emerald-700/60 hover:border-emerald-400" - > - Merge to base ↗ - </button> - )} - - <button - type="button" - onClick={() => navigate(`/exec/${taskId}`)} - className="px-2 py-1 font-mono text-[10px] uppercase tracking-wide text-[#9bc3ff] border border-[rgba(117,170,252,0.35)] hover:border-[#75aafc]" - > - Open in task page - </button> - </div> - - {/* Document body */} - <div - ref={containerRef} - onScroll={handleScroll} - className="flex-1 overflow-y-auto" - > - <div className="max-w-3xl mx-auto px-8 py-10 pb-32 text-[#dbe7ff]"> - <div className="flex items-center gap-3 mb-1"> - <h1 className="text-[24px] font-medium text-white tracking-tight"> - {label} - </h1> - {isStreaming && ( - <span className="flex items-center gap-1.5 px-2 py-0.5 bg-green-400/10 border border-green-400/30 text-green-400 font-mono text-[10px] uppercase"> - <span className="w-1.5 h-1.5 bg-green-400 rounded-full animate-pulse" /> - Live - </span> - )} - </div> - <p className="text-[10px] font-mono text-[#556677] uppercase tracking-wide mb-8"> - Live transcript — comments below are sent to the task as input. - </p> - - {loading && entries.length === 0 ? ( - <p className="text-[#556677] font-mono text-xs italic"> - Loading transcript… - </p> - ) : entries.length === 0 ? ( - <p className="text-[#556677] font-mono text-xs italic"> - {isStreaming ? "Waiting for output…" : "No output yet."} - </p> - ) : ( - <div className="space-y-4"> - {entries.map((entry, idx) => ( - <DocumentEntry key={idx} entry={entry} /> - ))} - {isStreaming && ( - <span className="inline-block w-2 h-4 bg-[#9bc3ff] animate-pulse align-baseline" /> - )} - </div> - )} - </div> - </div> - - {/* "Resume auto-scroll" floating chip when the user has scrolled up. */} - {showResumeScroll && ( - <button - type="button" - onClick={resumeScroll} - className="absolute bottom-32 right-6 z-10 px-3 py-1.5 font-mono text-[10px] uppercase tracking-wide text-[#9bc3ff] bg-[#091428] border border-[rgba(117,170,252,0.4)] hover:border-[#75aafc] shadow-lg" - > - ↓ Jump to latest - </button> - )} - - {/* Sticky comment composer — always pinned to the viewport bottom so - users can interact with the task no matter where they've scrolled. */} - <div - ref={composerRef} - className="absolute bottom-0 left-0 right-0 border-t border-dashed border-[rgba(117,170,252,0.25)] bg-[#091428]/95 backdrop-blur" - > - {sendError && ( - <div className="px-6 py-1 bg-red-900/20 text-red-400 text-xs font-mono"> - {sendError} - </div> - )} - <form - onSubmit={submitComment} - className="max-w-3xl mx-auto px-8 py-3 flex items-start gap-3" - > - <span className="text-[10px] font-mono text-[#556677] uppercase tracking-wide pt-2 shrink-0"> - Comment - </span> - <textarea - value={comment} - onChange={(e) => setComment(e.target.value)} - onKeyDown={(e) => { - // ⌘/Ctrl-Enter submits. - if ((e.metaKey || e.ctrlKey) && e.key === "Enter") { - void submitComment(e as unknown as React.FormEvent); - } - }} - placeholder={ - isStreaming - ? "Add a comment to interrupt and redirect…" - : "Task is not streaming — comments will queue if accepted." - } - rows={2} - disabled={sending} - className="flex-1 bg-transparent border border-[rgba(117,170,252,0.2)] focus:border-[#75aafc] outline-none px-3 py-2 text-[13px] text-[#dbe7ff] placeholder-[#445566] resize-none" - /> - <button - type="submit" - disabled={sending || !comment.trim()} - className="px-3 py-1.5 font-mono text-[10px] uppercase tracking-wide text-emerald-300 border border-emerald-700/60 hover:border-emerald-400 disabled:opacity-40 disabled:cursor-not-allowed shrink-0" - > - {sending ? "Sending…" : "Send"} - </button> - </form> - </div> - </div> - ); -} - -// --------------------------------------------------------------------------- -// Entry rendering — document-style, not log-style. -// --------------------------------------------------------------------------- - -function DocumentEntry({ entry }: { entry: TaskOutputEvent }) { - switch (entry.messageType) { - case "user_input": - return ( - <blockquote className="border-l-2 border-cyan-400/60 pl-4 py-1 italic text-cyan-200"> - <span className="not-italic text-[10px] font-mono text-cyan-400 uppercase tracking-wide block mb-1"> - You - </span> - {entry.content} - </blockquote> - ); - - case "assistant": - return ( - <div className="leading-relaxed text-[14px]"> - <SimpleMarkdown content={entry.content} className="text-[#e0eaf8]" /> - </div> - ); - - case "system": - return ( - <p className="text-[10px] font-mono text-[#556677] uppercase tracking-wide"> - {entry.content} - </p> - ); - - case "tool_use": - return ( - <p className="text-[11px] font-mono text-[#7788aa] flex items-center gap-2"> - <span className="text-yellow-500">·</span> - <span className="text-[#75aafc]">{entry.toolName || "tool"}</span> - {firstLineOfInput(entry.toolInput) && ( - <span className="text-[#445566] truncate"> - {firstLineOfInput(entry.toolInput)} - </span> - )} - </p> - ); - - case "tool_result": - if (!entry.content) return null; - return ( - <p className="text-[11px] font-mono pl-4"> - <span className={entry.isError ? "text-red-400" : "text-emerald-400"}> - {entry.isError ? "✗" : "→"} - </span>{" "} - <span className="text-[#7788aa]"> - {entry.content.split("\n")[0]} - {entry.content.includes("\n") && "…"} - </span> - </p> - ); - - case "result": - return ( - <div className="border-t border-[rgba(117,170,252,0.15)] pt-3 mt-6"> - <p className="text-[10px] font-mono text-emerald-400 uppercase tracking-wide mb-2"> - Result - </p> - <div className="leading-relaxed text-[13px]"> - <SimpleMarkdown content={entry.content} className="text-[#e0eaf8]" /> - </div> - {(entry.costUsd !== undefined || entry.durationMs !== undefined) && ( - <p className="text-[10px] font-mono text-[#556677] mt-2"> - {entry.durationMs !== undefined && - `Duration: ${(entry.durationMs / 1000).toFixed(1)}s`} - {entry.costUsd !== undefined && entry.durationMs !== undefined && " · "} - {entry.costUsd !== undefined && - `Cost: $${entry.costUsd.toFixed(4)}`} - </p> - )} - </div> - ); - - case "error": - return ( - <p className="border-l-2 border-red-400/60 pl-4 py-1 text-red-300 text-[13px]"> - {entry.content} - </p> - ); - - default: - // Fall back to a quiet rendering for unknown message types so users - // still see the data, just inconspicuously. - if (!entry.content) return null; - return ( - <p className="text-[11px] font-mono text-[#556677]"> - {entry.content} - </p> - ); - } -} - -/** Terminal task statuses where the merge button is meaningful. */ -function isTerminalStatus(status?: string): boolean { - if (!status) return false; - return ["done", "completed", "merged"].includes(status); -} - -function firstLineOfInput(input?: Record<string, unknown>): string { - if (!input) return ""; - // Common shapes — show the most informative single value. - for (const key of ["command", "file_path", "path", "url", "pattern", "query"]) { - const v = input[key]; - if (typeof v === "string" && v.length > 0) { - return v.split("\n")[0].slice(0, 96); - } - } - return ""; -} |
