@marcoscale98/pi-extensible-workflows 5.17.0-fork.1 → 6.2.1-fork.1

package/dist/trajectory/src/assets/index.html

@@ -142,7 +142,7 @@
     .workflow-name { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; font: 600 16px/1.2 var(--serif); }
     .workflow-meta { color: var(--dim); font: 11px ui-monospace, Menlo, Consolas, monospace; white-space: nowrap; }
     .agent-grid { min-width: 900px; padding: 0 12px 8px 28px; }
-    .agent-grid-head, .agent-grid-row { display: grid; grid-template-columns: minmax(220px, 2fr) minmax(140px, 1.2fr) 80px minmax(150px, 1.5fr) 80px 72px 64px; gap: 8px; align-items: center; }
+    .agent-grid-head, .agent-grid-row { display: grid; grid-template-columns: minmax(220px, 2fr) minmax(140px, 1.2fr) minmax(150px, 1.5fr) 80px 72px 64px; gap: 8px; align-items: center; }
     .agent-grid-head { padding: 4px 8px; color: var(--mute); font-size: 10px; letter-spacing: .05em; }
     .agent-grid-row { min-height: 24px; padding: 3px 8px; cursor: pointer; font-size: 11px; border-radius: 8px; }
     .agent-grid-row:hover { background: var(--bg-hover); }
@@ -157,6 +157,14 @@
     .log-row { display: grid; grid-template-columns: 7.5rem minmax(0, 1fr); gap: 10px; padding: 5px 8px; border-top: 1px solid var(--line); font: 11px ui-monospace, Menlo, Consolas, monospace; }
     .log-time { color: var(--mute); white-space: nowrap; }
     .log-message { min-width: 0; white-space: pre-wrap; overflow-wrap: anywhere; }
+    .calls-list { max-height: 480px; overflow: auto; }
+    .call-row { display: grid; grid-template-columns: 16px 7.5rem minmax(80px, 160px) minmax(0, 1fr) 64px; gap: 10px; align-items: center; padding: 5px 8px; border-top: 1px solid var(--line); font: 11px ui-monospace, Menlo, Consolas, monospace; cursor: pointer; list-style: none; }
+    .call-row::-webkit-details-marker { display: none; }
+    .call-row:hover { background: var(--bg-hover); }
+    .call-row b, .call-preview { min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+    .call-preview { color: var(--dim); }
+    .call-row .num { color: var(--dim); text-align: right; }
+    .call-body { padding: 4px 8px 12px 34px; }
     .row { display: grid; grid-template-columns: 16px 1fr; gap: 8px; align-items: center; padding: 3px 12px; cursor: pointer; }
     .row.child { padding-left: 28px; }
     .row:hover { background: var(--bg-hover); }
@@ -341,7 +349,7 @@
   <script src="./morphdom.min.js"></script>
   <script src="./prism.min.js"></script>
   <script>
-    const defaultRunLayout = () => ({ swimHeight: 220, ganttCollapsed: false, agentsCollapsed: false, logsCollapsed: false });
+    const defaultRunLayout = () => ({ swimHeight: 220, ganttCollapsed: false, agentsCollapsed: false, callsCollapsed: false, logsCollapsed: false });
     const maxSwimHeight = () => Math.max(72, window.innerHeight * 0.7);
     function loadRunLayout() {
       const fallback = defaultRunLayout();
@@ -354,6 +362,7 @@
           swimHeight: Number.isFinite(value.swimHeight) && value.swimHeight >= 72 ? value.swimHeight : fallback.swimHeight,
           ganttCollapsed: typeof value.ganttCollapsed === "boolean" ? value.ganttCollapsed : fallback.ganttCollapsed,
           agentsCollapsed: typeof value.agentsCollapsed === "boolean" ? value.agentsCollapsed : fallback.agentsCollapsed,
+          callsCollapsed: typeof value.callsCollapsed === "boolean" ? value.callsCollapsed : fallback.callsCollapsed,
           logsCollapsed: typeof value.logsCollapsed === "boolean" ? value.logsCollapsed : fallback.logsCollapsed,
         };
       } catch {
@@ -372,7 +381,7 @@
       }
     }
     function saveSidebarCollapsed() { try { localStorage.setItem("traj-sidebar-collapsed", JSON.stringify([...state.sidebarCollapsed])); } catch { /* Preferences are optional. */ } }
-    const state = { publishers: [], currentPub: null, currentRun: null, currentTarget: null, currentAgent: null, pendingTarget: null, agentRange: null, agentRangeAgent: null, selectedEvent: null, timelineHighlight: null, eventsSig: null, toolPane: "summary", sysPane: "prompt", inspMode: "event", inspSig: null, socket: null, transcripts: {}, transcriptPending: {}, transcriptSources: {}, transcriptRefresh: {}, transcriptRevision: {}, transcriptStatus: {}, transcriptCacheOrder: [], actionTargets: new Map(), shareUrls: {}, runLayout: loadRunLayout(), sidebarCollapsed: loadSidebarCollapsed(), workflowArgumentsOpen: new Set() };
+    const state = { publishers: [], currentPub: null, currentRun: null, currentTarget: null, currentAgent: null, pendingTarget: null, agentRange: null, agentRangeAgent: null, selectedEvent: null, timelineHighlight: null, eventsSig: null, toolPane: "summary", sysPane: "prompt", inspMode: "event", inspSig: null, socket: null, transcripts: {}, transcriptPending: {}, transcriptSources: {}, transcriptRefresh: {}, transcriptRevision: {}, transcriptStatus: {}, transcriptCacheOrder: [], actionTargets: new Map(), shareUrls: {}, runLayout: loadRunLayout(), sidebarCollapsed: loadSidebarCollapsed(), workflowArgumentsOpen: new Set(), callsOpen: new Set() };
     let hasAcceptedState = false;
     let socketGeneration = 0;
     let socketReconnect;
@@ -676,6 +685,8 @@
         const style = running ? `left:${Math.max(0, left / span * 100)}%;width:0%` : barStyle(Math.max(0, left), agent.durationMs || 0, 0, span);
         lanes.push({ key: `agent:${agent.id}`, html: `<div class="lane${agent.parentId ? " child" : ""}" data-agent="${esc(agent.id)}">${glyph(agent.state)}<div class="lab">${esc(agent.label || agent.name)}</div><div class="track"><i class="bar ${stateClass(agent.state)}" style="${style}"${running ? ` data-clock-bar-start="${agent.startedAt || start}" data-clock-bar-min="2"` : ""}></i>${toolBars}</div></div>` });
       }
+      const calls = scriptCalls(record);
+      if (calls.length) lanes.push({ key: "calls", html: `<div class="lane crumb"><span></span><div class="lab">script</div><div class="track">${calls.map((call) => { const status = callState(record, call); const running = status === "running"; const title = `${callLabel(call)} · ${status}${running ? "" : ` · ${callDuration(call)}`}`; return `<i class="bar tool ${stateClass(status)}" style="${running ? `left:${Math.max(0, (call.startedAt - start) / span * 100)}%;width:0.5%` : barStyle(call.startedAt - start, (call.finishedAt ?? call.startedAt) - call.startedAt, 0.5, span)}"${running ? ` data-clock-bar-start="${call.startedAt}" data-clock-bar-min="0.5"` : ""} title="${esc(title)}"></i>`; }).join("")}</div></div>` });
       return [{ key: "axis", html: `<div class="axis"><div></div><div class="ticks"><span>${fmtClock(start)}</span><span data-clock-axis="mid">${live ? "" : `+${fmtRuntime(span / 2)}`}</span><span data-clock-axis="end">${live ? "" : "end"}</span></div></div>` }, ...lanes];
     }
     // Only the lanes whose markup changed are re-parsed; the quantised span keeps the others byte-identical between frames.
@@ -721,15 +732,26 @@
         const phaseAgents = phase.agents.flatMap((id) => { const agent = byId.get(id); return agent ? [agent] : []; });
         const groups = agentGridGroups(phaseAgents);
         const showScopes = groups.length > 1 || groups.some(({ label }) => label !== "Agents");
-        return `<div class="phaseband">PHASE ${esc(phase.name)}</div>${groups.map((group) => `${showScopes ? `<div class="agent-grid-scope">${esc(group.label)}</div>` : ""}${group.agents.map((agent) => { const depth = agentDepth(agent, byId); const label = agent.label || agent.name; const settled = depth ? "•" : "▾"; const indent = (showScopes ? 12 : 0) + depth * 12; const toolSummary = toolTimingSummary(timingsByAgent.get(agent.id) || []); return `<div class="agent-grid-row${agent.id === state.currentAgent ? " on" : ""}" data-agent="${esc(agent.id)}"><div class="agent-cell" style="padding-left:${indent}px">${settled} ${esc(label)}</div><div class="agent-cell">${esc(agentModel(agent))}</div><div class="agent-cell">${esc(agent.role || "agent")}</div><div class="agent-cell" title="${esc(toolSummary)}">${esc(toolSummary)}</div><div class="agent-cell num">${fmtRuntime(agent.durationMs || 0)}</div><div class="agent-cell num">${fmtTokens(accounting(agent).input + accounting(agent).output)}</div><div class="agent-cell num">${fmtCost(accounting(agent).cost)}</div></div>`; }).join("")}`).join("")}`;
+        return `<div class="phaseband">PHASE ${esc(phase.name)}</div>${groups.map((group) => `${showScopes ? `<div class="agent-grid-scope">${esc(group.label)}</div>` : ""}${group.agents.map((agent) => { const depth = agentDepth(agent, byId); const label = agent.label || agent.name; const settled = depth ? "•" : "▾"; const indent = (showScopes ? 12 : 0) + depth * 12; const toolSummary = toolTimingSummary(timingsByAgent.get(agent.id) || []); return `<div class="agent-grid-row${agent.id === state.currentAgent ? " on" : ""}" data-agent="${esc(agent.id)}"><div class="agent-cell" style="padding-left:${indent}px">${settled} ${esc(label)}</div><div class="agent-cell">${esc(agentModel(agent))}</div><div class="agent-cell" title="${esc(toolSummary)}">${esc(toolSummary)}</div><div class="agent-cell num">${fmtRuntime(agent.durationMs || 0)}</div><div class="agent-cell num">${fmtTokens(accounting(agent).input + accounting(agent).output)}</div><div class="agent-cell num">${fmtCost(accounting(agent).cost)}</div></div>`; }).join("")}`).join("")}`;
       }).join("");
-      return `<div class="agent-grid"><div class="agent-grid-head"><span>AGENT</span><span>MODEL</span><span>ROLE</span><span>TOOL TIMING</span><span>DURATION</span><span>TOKENS</span><span>COST</span></div>${body}</div>`;
+      return `<div class="agent-grid"><div class="agent-grid-head"><span>AGENT</span><span>MODEL</span><span>TOOL TIMING</span><span>DURATION</span><span>TOKENS</span><span>COST</span></div>${body}</div>`;
     }
     function renderLogs(record) { const events = record.run.events || []; const logs = events.filter((event) => event.type === "log"); const omitted = events.find((event) => event.type === "trajectory:truncated"); if (!logs.length && !omitted) return ""; const collapsed = state.runLayout.logsCollapsed; return `<section class="logs"><button type="button" class="sec section-toggle" data-toggle="logs" aria-expanded="${!collapsed}">${collapsed ? "▸" : "▾"} LOGS</button>${collapsed ? "" : `<div class="logs-list">${logs.map((event) => `<div class="log-row"><span class="log-time">${esc(fmtClock(event.timestamp))}</span><span class="log-message">${esc(event.message)}</span></div>`).join("")}${omitted ? `<div class="log-row"><span class="log-time">—</span><span class="log-message">${esc(omitted.omitted)} events omitted from live state</span></div>` : ""}</div>`}</section>`; }
+    const scriptCalls = (record) => (record.run.scriptCalls || []).filter((call) => call?.type !== "trajectory:truncated");
+    // The journal proves completion even if a crash prevented the finish metadata write.
+    const callState = (record, call) => call.output?.status === "completed" || call.output?.status === "available" || call.output?.status === "truncated" && call.output.kind === "result" ? "completed" : call.error ? (call.error.code === "CANCELLED" ? "stopped" : "failed") : call.finishedAt !== undefined ? "completed" : record.run.state === "running" || record.run.state === "pausing" ? "running" : "interrupted";
+    const callDuration = (call) => { if (call.finishedAt === undefined) return "—"; const ms = Math.max(0, call.finishedAt - call.startedAt); return ms < 1000 ? `${ms}ms` : fmtRuntime(ms); };
+    const callLabel = (call) => call.kind === "shell" ? "shell" : `tools.${call.name}`;
+    const callPreview = (call) => call.kind === "tool" ? call.error?.code || "" : typeof call.input === "string" ? call.input : JSON.stringify(call.input);
+    function renderCalls(record) {
+      const calls = scriptCalls(record); const omitted = (record.run.scriptCalls || []).find((call) => call?.type === "trajectory:truncated"); if (!calls.length && !omitted) return ""; const collapsed = state.runLayout.callsCollapsed;
+      const rows = calls.map((call) => { const key = `${state.currentRun}\t${call.path}`; const status = callState(record, call); const duration = status === "running" ? `<span data-clock-kind="runtime" data-clock-start="${call.startedAt}" data-clock-end="live"></span>` : call.finishedAt === undefined ? "—" : esc(callDuration(call)); const input = call.kind !== "shell" ? "" : call.inputBytes !== undefined ? `<div class="hint">Input truncated (${esc(call.inputBytes)} bytes).</div><div class="code">${esc(call.input)}</div>` : `<div class="code">${highlightCode(json(call.input), "json")}</div>`; return `<details class="call"${state.callsOpen.has(key) ? " open" : ""} data-call="${esc(key)}"><summary class="call-row">${glyph(status)}<span class="log-time">${esc(fmtClock(call.startedAt))}</span><b>${esc(callLabel(call))}</b><span class="call-preview">${esc(callPreview(call))}</span><span class="num">${duration}</span></summary><div class="call-body">${call.phase ? `<div class="kv"><div class="k">Phase</div><div>${esc(call.phase)}</div></div>` : ""}${call.kind === "shell" ? `<div class="sec">INPUT</div>${input}<div class="sec">OUTPUT</div>${renderOutputPane(call)}` : `<div class="kv"><div class="k">Status</div><div>${esc(status)}</div>${call.error ? `<div class="k">Error</div><div>${esc(call.error.code)}</div>` : ""}</div>`}</div></details>`; }).join("");
+      return `<section class="logs"><button type="button" class="sec section-toggle" data-toggle="calls" aria-expanded="${!collapsed}">${collapsed ? "▸" : "▾"} CALLS</button>${collapsed ? "" : `<div class="calls-list">${rows}${omitted ? `<div class="log-row"><span class="log-time">—</span><span class="log-message">${esc(omitted.omitted)} calls omitted from live state</span></div>` : ""}</div>`}</section>`;
+    }
     function renderAgentSection(record, timingsByAgent) { const collapsed = state.runLayout.agentsCollapsed; return `<button type="button" class="sec section-toggle" data-toggle="agents" aria-expanded="${!collapsed}">${collapsed ? "▸" : "▾"} AGENTS</button>${collapsed ? "" : renderAgentGrid(record, timingsByAgent)}`; }
-    function renderList(record, timingsByAgent) { const run = record.run; const completed = run.agents.filter((agent) => agent.state === "completed").length; return `<section class="workflow-card"><button type="button" class="workflow-head on" data-run="${esc(state.currentRun)}">${glyph(run.state)}<span class="workflow-name">${esc(run.workflowName)}</span><span class="workflow-meta">${completed}/${run.agents.length} agents · ${fmtCost(runCost(record))} · <span data-clock-kind="runtime" data-clock-start="${startOf(record).getTime()}" data-clock-end="${run.state === "running" ? "live" : endOf(record)}"></span></span></button>${renderAgentSection(record, timingsByAgent)}${renderLogs(record)}</section>`; }
+    function renderList(record, timingsByAgent) { const run = record.run; const completed = run.agents.filter((agent) => agent.state === "completed").length; return `<section class="workflow-card"><button type="button" class="workflow-head on" data-run="${esc(state.currentRun)}">${glyph(run.state)}<span class="workflow-name">${esc(run.workflowName)}</span><span class="workflow-meta">${completed}/${run.agents.length} agents · ${fmtCost(runCost(record))} · <span data-clock-kind="runtime" data-clock-start="${startOf(record).getTime()}" data-clock-end="${run.state === "running" ? "live" : endOf(record)}"></span></span></button>${renderAgentSection(record, timingsByAgent)}${renderCalls(record)}${renderLogs(record)}</section>`; }
     function actionButton(label, action, disabled) { return `<button class="btn-alt" data-act="${action}"${disabled ? " disabled" : ""}>${label}</button>`; }
-    function subagentAgent(subagent) { return { id: subagent.id, name: subagentLabel(subagent), label: subagentLabel(subagent), role: subagent.role || "subagent", state: subagent.state, model: subagent.model, requestedModel: subagent.request.model, tools: subagent.tools || [], toolDefinitions: subagent.toolDefinitions || [], skills: subagent.request.skills || [], extensions: subagent.request.extensions || [], contextFiles: subagent.request.contextFiles || [], prompt: subagent.request.prompt, systemPrompt: subagent.attempt?.setup?.systemPrompt || subagent.request.prompt, attempts: subagent.attempts || 0, mode: subagent.mode, cwd: subagent.cwd, worktree: subagent.worktree, output: subagent.output || subagentOutputFallback(subagent), outcome: subagent.result !== undefined ? { kind: "result", value: subagent.result } : subagent.failure !== undefined ? { kind: "failure", value: subagent.failure } : subagent.attempt?.error !== undefined ? { kind: "failure", value: subagent.attempt.error } : undefined, finishedAt: subagent.finishedAt, attemptDetails: subagent.attempt ? [subagent.attempt] : [] }; }
+    function subagentAgent(subagent) { return { id: subagent.id, name: subagentLabel(subagent), label: subagentLabel(subagent), state: subagent.state, model: subagent.model, requestedModel: subagent.request.model, tools: subagent.tools || [], toolDefinitions: subagent.toolDefinitions || [], skills: subagent.request.skills || [], extensions: subagent.request.extensions || [], contextFiles: subagent.request.contextFiles || [], prompt: subagent.request.prompt, systemPrompt: subagent.attempt?.setup?.systemPrompt || subagent.request.prompt, attempts: subagent.attempts || 0, mode: subagent.mode, cwd: subagent.cwd, worktree: subagent.worktree, output: subagent.output || subagentOutputFallback(subagent), outcome: subagent.result !== undefined ? { kind: "result", value: subagent.result } : subagent.failure !== undefined ? { kind: "failure", value: subagent.failure } : subagent.attempt?.error !== undefined ? { kind: "failure", value: subagent.attempt.error } : undefined, finishedAt: subagent.finishedAt, attemptDetails: subagent.attempt ? [subagent.attempt] : [] }; }
     function subagentOutputFallback(subagent) { if (subagent.result !== undefined) return { status: "available", value: subagent.result, bytes: 0 }; if (subagent.failure !== undefined) return { status: "failed", code: subagent.failure.code || "AGENT_FAILED", message: subagent.failure.message || "Subagent failed" }; if (subagent.state === "running") return { status: "pending" }; if (subagent.state === "completed") return { status: "unavailable" }; if (subagent.state === "stopped") return { status: "cancelled", code: subagent.error?.code || "CANCELLED", message: subagent.error?.message || "Subagent stopped" }; if (subagent.state === "failed") return { status: "failed", code: subagent.error?.code || "AGENT_FAILED", message: subagent.error?.message || "Subagent failed" }; return { status: "unavailable" }; }
     function renderAgentStats(found, agent, isSubagent) {
       const offline = !found.publisher.connected; const stateText = agent.state || "—"; const value = isSubagent ? subagentAccounting(found.record) : accounting(agent);
@@ -739,7 +761,7 @@
       return `${glyph(stateText)} <button type="button" class="btn-alt agent-details" data-agent-details="1">Agent details</button> ${fmtCost(value.cost)} · ${fmtTokens(value.input + value.output)} · runtime ${clock}${controls ? `<span class="actions">${controls}</span>` : ""}${isSubagent && offline ? '<span class="offline">publisher offline — actions disabled</span>' : ""}`;
     }
     function renderWorkflowArguments(record, open = false, key = "") { const args = record.snapshot?.args; const incomplete = record.snapshotArgsTruncated === true; if (args === null || args === undefined && !incomplete) return ""; const status = incomplete ? ' <span class="ins-sub">incomplete</span>' : ""; const body = args === undefined ? '<div class="hint">Arguments unavailable in the live transport.</div>' : `<div class="code args">${highlightCode(json(args), "json")}</div>`; const warning = incomplete ? '<div class="workflow-args-warning">Arguments were truncated by the live transport; this value is incomplete.</div>' : ""; return `<details class="workflow-args"${open ? " open" : ""} data-workflow-args="${esc(key)}"><summary class="sec">ARGUMENTS${status}</summary>${body}${warning}</details>`; }
-    function renderDossier(publisher, record) { const offline = !publisher.connected; const issues = record.run.agents.filter((agent) => agent.state === "failed").map((agent) => `<div class="pending issue"><div class="ttl">${glyph(agent.state)} ${esc(agent.label || agent.name)} needs retry</div><div class="ins-sub">${esc(agent.attemptDetails?.at(-1)?.error?.message || record.run.error?.message || "agent failed")}</div><div class="actions">${actionButton("Retry", "retry", offline)}</div>${offline ? '<div class="offline">publisher offline — actions disabled</div>' : ""}</div>`).join(""); const pending = issues + (record.awaiting || []).map((item) => `<div class="pending" data-name="${esc(item.name)}"><div class="ttl">${glyph("awaiting_input")} checkpoint ${esc(item.name)}</div><div class="ins-sub">${esc(item.prompt)}</div><div class="actions">${actionButton("Approve", "checkpoint-approve", offline)}${actionButton("Reject", "checkpoint-reject", offline)}</div>${offline ? '<div class="offline">publisher offline — actions disabled</div>' : ""}</div>`).join(""); const run = record.run; const controls = (run.state === "running" || run.state === "awaiting_input" ? `${actionButton("Pause", "pause", offline)}${actionButton("Abort run", "stop", offline)}` : run.state === "pausing" ? `${actionButton("Cancel pause", "resume", offline)}${actionButton("Abort run", "stop", offline)}` : run.state === "paused" || run.state === "interrupted" || run.state === "budget_exhausted" ? `${actionButton("Resume", "resume", offline)}${actionButton("Abort run", "stop", offline)}` : "") + (window.__PIEWF_STATIC__ ? "" : actionButton("Share", "share", offline)); const rows = run.agents.map((agent) => `<tr data-agent="${esc(agent.id)}"><td>${glyph(agent.state)}</td><td><b>${esc(agent.label || agent.name)}</b><div class="ins-sub">${esc(agent.role || "agent")}</div></td><td class="num">${fmtTokens(accounting(agent).input + accounting(agent).output)}</td><td class="num">${fmtCost(accounting(agent).cost)}</td></tr>`).join(""); const script = record.snapshot?.script ? `<div class="sec">SCRIPT <span class="ins-sub">click to expand</span></div><div class="code script" data-script="1">${highlightCode(record.snapshot.script, "javascript")}</div>` : ""; const argumentsSection = renderWorkflowArguments(record, state.workflowArgumentsOpen.has(state.currentRun), state.currentRun); return `<div class="ins-head" data-run-panel="1">${glyph(run.state)}<b>${esc(run.workflowName)}</b><span class="ins-sub">${esc(run.state)}</span></div><div class="ins-body">${pending ? '<div class="sec">NEEDS ATTENTION</div>' + pending : ""}<div class="sec">STATE</div><div class="kv"><div class="k">Phase</div><div>${esc(run.phase || "—")}</div><div class="k">Active</div><div>${run.agents.filter((agent) => agent.state === "running").length} running · ${run.agents.filter((agent) => agent.state === "queued").length} queued · ${run.agents.filter((agent) => agent.state === "failed").length} failed</div><div class="k">Outcome</div><div>${esc(run.error ? `${run.error.code}: ${run.error.message}` : run.state === "completed" ? "completed" : attention(record) ? `${attention(record)} agent(s) need a human` : "—")}</div><div class="k">Cost</div><div>${fmtCost(runCost(record))}${run.budget?.costUsd?.hard !== undefined ? ` / ${fmtCost(run.budget.costUsd.hard)} hard` : ""}</div><div class="k">Tokens</div><div>${fmtTokens(runTokens(record))}${run.budget?.tokens?.hard !== undefined ? ` / ${fmtTokens(run.budget.tokens.hard)} hard` : ""}</div><div class="k">Started</div><div>${fmtClock(startOf(record))}</div><div class="k">Runtime</div><div data-clock-kind="runtime" data-clock-start="${startOf(record).getTime()}" data-clock-end="${run.state === "running" ? "live" : endOf(record)}"></div></div><div class="actions">${controls}</div>${state.shareUrls[run.id] ? `<div class="share-url"><input readonly value="${esc(state.shareUrls[run.id])}" /><button class="btn-alt" data-copy-share="${esc(state.shareUrls[run.id])}">Copy</button></div><div class="ins-sub">secret gist: anyone with the link can read it</div>` : ""}${script}${argumentsSection}<div class="sec">AGENTS</div><table class="acct"><thead><tr><th></th><th>agent</th><th class="num">tok</th><th class="num">cost</th></tr></thead><tbody>${rows}</tbody></table></div>`; }
+    function renderDossier(publisher, record) { const offline = !publisher.connected; const issues = record.run.agents.filter((agent) => agent.state === "failed").map((agent) => `<div class="pending issue"><div class="ttl">${glyph(agent.state)} ${esc(agent.label || agent.name)} needs retry</div><div class="ins-sub">${esc(agent.attemptDetails?.at(-1)?.error?.message || record.run.error?.message || "agent failed")}</div><div class="actions">${actionButton("Retry", "retry", offline)}</div>${offline ? '<div class="offline">publisher offline — actions disabled</div>' : ""}</div>`).join(""); const pending = issues + (record.awaiting || []).map((item) => `<div class="pending" data-name="${esc(item.name)}"><div class="ttl">${glyph("awaiting_input")} checkpoint ${esc(item.name)}</div><div class="ins-sub">${esc(item.prompt)}</div><div class="actions">${actionButton("Approve", "checkpoint-approve", offline)}${actionButton("Reject", "checkpoint-reject", offline)}</div>${offline ? '<div class="offline">publisher offline — actions disabled</div>' : ""}</div>`).join(""); const run = record.run; const controls = (run.state === "running" || run.state === "awaiting_input" ? `${actionButton("Pause", "pause", offline)}${actionButton("Abort run", "stop", offline)}` : run.state === "pausing" ? `${actionButton("Cancel pause", "resume", offline)}${actionButton("Abort run", "stop", offline)}` : run.state === "paused" || run.state === "interrupted" || run.state === "budget_exhausted" ? `${actionButton("Resume", "resume", offline)}${actionButton("Abort run", "stop", offline)}` : "") + (window.__PIEWF_STATIC__ ? "" : actionButton("Share", "share", offline)); const rows = run.agents.map((agent) => `<tr data-agent="${esc(agent.id)}"><td>${glyph(agent.state)}</td><td><b>${esc(agent.label || agent.name)}</b></td><td class="num">${fmtTokens(accounting(agent).input + accounting(agent).output)}</td><td class="num">${fmtCost(accounting(agent).cost)}</td></tr>`).join(""); const script = record.snapshot?.script ? `<div class="sec">SCRIPT <span class="ins-sub">click to expand</span></div><div class="code script" data-script="1">${highlightCode(record.snapshot.script, "javascript")}</div>` : ""; const argumentsSection = renderWorkflowArguments(record, state.workflowArgumentsOpen.has(state.currentRun), state.currentRun); return `<div class="ins-head" data-run-panel="1">${glyph(run.state)}<b>${esc(run.workflowName)}</b><span class="ins-sub">${esc(run.state)}</span></div><div class="ins-body">${pending ? '<div class="sec">NEEDS ATTENTION</div>' + pending : ""}<div class="sec">STATE</div><div class="kv"><div class="k">Phase</div><div>${esc(run.phase || "—")}</div><div class="k">Active</div><div>${run.agents.filter((agent) => agent.state === "running").length} running · ${run.agents.filter((agent) => agent.state === "queued").length} queued · ${run.agents.filter((agent) => agent.state === "failed").length} failed</div><div class="k">Outcome</div><div>${esc(run.error ? `${run.error.code}: ${run.error.message}` : run.state === "completed" ? "completed" : attention(record) ? `${attention(record)} agent(s) need a human` : "—")}</div><div class="k">Cost</div><div>${fmtCost(runCost(record))}${run.budget?.costUsd?.hard !== undefined ? ` / ${fmtCost(run.budget.costUsd.hard)} hard` : ""}</div><div class="k">Tokens</div><div>${fmtTokens(runTokens(record))}${run.budget?.tokens?.hard !== undefined ? ` / ${fmtTokens(run.budget.tokens.hard)} hard` : ""}</div><div class="k">Started</div><div>${fmtClock(startOf(record))}</div><div class="k">Runtime</div><div data-clock-kind="runtime" data-clock-start="${startOf(record).getTime()}" data-clock-end="${run.state === "running" ? "live" : endOf(record)}"></div></div><div class="actions">${controls}</div>${state.shareUrls[run.id] ? `<div class="share-url"><input readonly value="${esc(state.shareUrls[run.id])}" /><button class="btn-alt" data-copy-share="${esc(state.shareUrls[run.id])}">Copy</button></div><div class="ins-sub">secret gist: anyone with the link can read it</div>` : ""}${script}${argumentsSection}<div class="sec">AGENTS</div><table class="acct"><thead><tr><th></th><th>agent</th><th class="num">tok</th><th class="num">cost</th></tr></thead><tbody>${rows}</tbody></table></div>`; }
     function renderRunLayout() { const ganttCollapsed = state.runLayout.ganttCollapsed; const splitSwim = $("split-swim"); $("swim").classList.toggle("hidden", ganttCollapsed); splitSwim.classList.toggle("hidden", ganttCollapsed); const heading = $("toggle-gantt"); heading.textContent = `${ganttCollapsed ? "▸" : "▾"} GANTT`; heading.setAttribute("aria-expanded", String(!ganttCollapsed)); const swimHeight = Math.min(state.runLayout.swimHeight, maxSwimHeight()); document.documentElement.style.setProperty("--swim", `${swimHeight}px`); }
     function clampSwimHeight() { renderRunLayout(); }
     function renderHome() { const homePublisher = livePublishers().find((publisher) => publisher.id === state.currentPub); const hasSubagents = Boolean(homePublisher?.subagents?.length); $("canvas-pub").textContent = homePublisher ? homePublisher.id.slice(0, 8) : "—"; $("canvas-run").textContent = hasSubagents ? "SUBAGENTS" : "—"; renderedRunHtml.clear(); ganttOrder = ""; patch($("swim-content"), hasSubagents ? renderSubagentGantt(homePublisher) : ""); $("swim").dataset.timelineHasTime = hasSubagents ? "true" : "false"; $("swim").dataset.timelineRunning = String(Boolean(hasSubagents && homePublisher.subagents.some((subagent) => subagent.state === "running"))); if (hasSubagents) { const subagents = homePublisher.subagents || []; const start = Math.min(...subagents.map((subagent) => subagentStartOf(subagent))); const end = Math.max(start + 1, ...subagents.map((subagent) => subagentEndOf(subagent))); $("swim").dataset.timelineStart = String(start); $("swim").dataset.timelineSpan = String(Math.max(1, end - start)); $("swim").querySelector(":scope > .now")?.classList.remove("hidden"); } else { $("swim").querySelector(":scope > .now")?.classList.add("hidden"); } patch($("list"), hasSubagents ? '<div class="hint">Select a subagent to inspect its transcript.</div>' : ""); patch($("run-stats"), hasSubagents ? `${homePublisher.subagents.length} subagents` : ""); $("run-crumb").textContent = "—"; $("agent-crumb").textContent = "—"; patch($("agent-ticks"), ""); $("agent-timeline").dataset.timelineHasTime = "false"; $("agent-timeline").querySelector(":scope > .now")?.classList.add("hidden"); $("agent-timeline").querySelector(":scope > .brush")?.classList.add("hidden"); patch($("events"), ""); $("q").value = ""; $("reset-agent-range").classList.add("hidden"); patch($("insp"), hasSubagents ? '<div class="ins-body hint">Select a subagent.</div>' : '<div class="ins-body hint">Select a run.</div>'); state.inspSig = null; renderRunLayout(); }
@@ -764,9 +786,9 @@
       const { skillFilters, skillResolved, extensionFilters, extensionResolved } = sessionResources(agent);
       const attempt = agent.attemptDetails?.at(-1);
       const attempts = (agent.attemptDetails || []).map((detail) => ({ attempt: detail.attempt, transport: detail.transport, cwd: detail.setup?.cwd, model: detail.setup?.model, tools: detail.setup?.tools, ...(detail.error ? { error: detail.error } : {}) }));
-      const environment = { role: agent.role, requestedModel: agent.requestedModel, model: agent.model, cwd: agent.cwd || attempt?.setup?.cwd || selected()?.record?.run?.cwd, mode: agent.mode, contextFiles: agent.contextFiles, worktree: agent.worktree || (agent.worktreeOwner ? { owner: agent.worktreeOwner } : undefined), structuralPath: agent.structuralPath, parentId: agent.parentId, parentBreadcrumb: agent.parentBreadcrumb, handle: agent.handle, turn: agent.turn, continuity: agent.continuity, state: agent.state, attempts: agent.attempts, attempt: attempt?.attempt, attemptDetails: attempts };
+      const environment = { requestedModel: agent.requestedModel, model: agent.model, cwd: agent.cwd || attempt?.setup?.cwd || selected()?.record?.run?.cwd, mode: agent.mode, contextFiles: agent.contextFiles, worktree: agent.worktree || (agent.worktreeOwner ? { owner: agent.worktreeOwner } : undefined), structuralPath: agent.structuralPath, parentId: agent.parentId, parentBreadcrumb: agent.parentBreadcrumb, handle: agent.handle, turn: agent.turn, continuity: agent.continuity, state: agent.state, attempts: agent.attempts, attempt: attempt?.attempt, attemptDetails: attempts };
       const panes = {
-        prompt: () => `<div class="kv"><div class="k">Tokens</div><div>est. ${estTokens(prompt)}</div></div><div class="md">${promptAvailable ? sanitizeMarkdown(marked.parse(prompt, { mangle: false, headerIds: false })) : "Prompt unavailable"}</div>`,
+        prompt: () => `<div class="kv"><div class="k">Tokens</div><div>est. ${estTokens(prompt)}</div></div><div class="code">${promptAvailable ? esc(prompt) : "Prompt unavailable"}</div>`,
         tools: () => (agent.tools || []).map((tool) => { const description = toolDescriptionOf(agent, tool); return `<div class="tool-item"><b>${esc(tool)}</b>${description ? `<span>${esc(description)}</span>` : ""}</div>`; }).join("") || '<div class="hint">No tools recorded.</div>',
         skills: () => `<div class="sec">FILTERS</div>${nameList(skillFilters)}<div class="sec">RESOLVED</div>${resourceList(skillResolved)}`,
         extensions: () => `<div class="sec">FILTERS</div>${nameList(extensionFilters)}<div class="sec">RESOLVED</div>${resourceList(extensionResolved)}`,
@@ -775,7 +797,7 @@
       };
       patch($("sys-pane"), (panes[state.sysPane] || panes.prompt)()); $("sys-tabs").querySelectorAll("span").forEach((tab) => tab.classList.toggle("on", tab.dataset.pane === state.sysPane));
     }
-    function renderSystemPromptPane(agent, entry) { const prompt = typeof entry?.prompt === "string" ? entry.prompt : typeof agent.systemPrompt === "string" ? agent.systemPrompt : ""; const available = typeof entry?.prompt === "string" || typeof agent.systemPrompt === "string"; patch($("sys-pane"), `<div class="kv"><div class="k">Tokens</div><div>est. ${estTokens(prompt)}</div></div><div class="md">${available ? sanitizeMarkdown(marked.parse(prompt, { mangle: false, headerIds: false })) : "System prompt unavailable"}</div>`); }
+    function renderSystemPromptPane(agent, entry) { const prompt = typeof entry?.prompt === "string" ? entry.prompt : typeof agent.systemPrompt === "string" ? agent.systemPrompt : ""; const available = typeof entry?.prompt === "string" || typeof agent.systemPrompt === "string"; patch($("sys-pane"), `<div class="kv"><div class="k">Tokens</div><div>est. ${estTokens(prompt)}</div></div><div class="code">${available ? esc(prompt) : "System prompt unavailable"}</div>`); }
     function renderAssistantToolCalls(entry, entries) { const toolCalls = toolCallsOf(entry); return toolCalls.length ? `<div class="assistant-tools">${toolCalls.map((call) => `<div class="assistant-tool"><span class="pill tool">${esc(call.name)}</span><div class="code">${highlightCode(json(toolArgsOf(call, entries)), "json")}</div></div>`).join("")}</div>` : ""; }
     function renderAgentDetails(agent, root = $("insp")) { patch(root, `<div class="ins-head"><span class="pill context">AGENT</span><b class="tool-name">${esc(agent.label || agent.name)}</b><span class="ins-sub">${esc(agent.state || "")}</span></div><div class="itabs" id="sys-tabs"><span data-pane="prompt">Prompt</span><span data-pane="tools">Tools</span><span data-pane="skills">Skills</span><span data-pane="extensions">Extensions</span><span data-pane="env">Environment</span><span data-pane="output">Output</span></div><div class="ins-body"><div id="sys-pane"></div></div>`); renderSystemPane(agent); }
     function renderInspector(detail, agent, entries, root = $("insp")) { if (!detail) { patch(root, '<div class="ins-body hint">Select an event.</div>'); return; } if (detail.kind === "tool" || detail.kind === "skill") { patch(root, `<div class="ins-head"><span class="pill ${detail.kind}">${eventLabel(detail.kind)}</span><b class="tool-name">${esc(toolNameOf(detail.entry))}</b><span class="ins-sub">${esc(detail.entry.timestamp || "")}</span></div><div class="itabs" id="tool-tabs"><span data-pane="summary">Summary</span><span data-pane="payload">Input</span><span data-pane="result">Output</span><span data-pane="schema">Schema</span><span data-pane="timing">Timing</span></div><div class="ins-body"><div id="tool-pane"></div></div>`); renderToolPane(detail, entries); return; } if (detail.kind === "system") { const prompt = typeof detail.entry.prompt === "string" ? detail.entry.prompt : agent.systemPrompt; patch(root, `<div class="ins-head"><span class="pill system">SYSTEM</span><span class="ins-sub">est. ${estTokens(prompt || "")} tok</span></div><div class="ins-body"><div id="sys-pane"></div></div>`); renderSystemPromptPane(agent, detail.entry); return; } const message = detail.entry.message || detail.entry; const body = detail.kind === "assistant" ? assistantBodyText(message.content ?? detail.entry.content) : contentText(message.content ?? detail.entry.content); const tokenRows = messageTokenRows(detail.kind, message, body); const bodyHtml = detail.kind === "assistant" ? sanitizeMarkdown(marked.parse(body)) : esc(body); const tokenHtml = tokenRows ? `<div class="kv">${tokenRows}</div>` : ""; const assistantTools = detail.kind === "assistant" ? renderAssistantToolCalls(detail.entry, entries) : ""; patch(root, `<div class="ins-head"><span class="pill ${detail.kind}">${eventLabel(detail.kind)}</span></div><div class="ins-body">${tokenHtml}<div class="${detail.kind === "assistant" ? "md" : "code"}">${bodyHtml}</div>${assistantTools}</div>`); }
@@ -862,8 +884,8 @@
       });
     }
     $("script-close").addEventListener("click", closeScript); $("script-bg").addEventListener("click", closeScript); document.addEventListener("keydown", (event) => { if (event.key === "Escape") { closeScript(); setAgentRange(null); if (document.body.dataset.view === "run") renderRun(); else renderAgent(); } });
-    document.addEventListener("toggle", (event) => { const target = event.target; if (!(target instanceof HTMLDetailsElement) || target.dataset.workflowArgs === undefined) return; if (target.open) state.workflowArgumentsOpen.add(target.dataset.workflowArgs); else state.workflowArgumentsOpen.delete(target.dataset.workflowArgs); }, true);
-    document.querySelector(".app").addEventListener("click", (event) => { const target = event.target.closest("[data-sidebar-group], [data-sidebar-section], [data-copy-share], [data-act], [data-run], [data-subagent], [data-agent], [data-agent-details], [data-script], [data-pane], [data-toggle]"); if (!target) return; if (target.dataset.copyShare) { const input = target.previousElementSibling; navigator.clipboard?.writeText(target.dataset.copyShare).then(() => showToast("copied"), () => { input?.select(); showToast("press Ctrl+C to copy"); }); return; } if (target.dataset.sidebarGroup !== undefined || target.dataset.sidebarSection !== undefined) { const key = target.dataset.sidebarGroup ?? target.dataset.sidebarSection; if (state.sidebarCollapsed.has(key)) state.sidebarCollapsed.delete(key); else state.sidebarCollapsed.add(key); saveSidebarCollapsed(); renderSidebar(); return; } if (target.dataset.toggle) { if (document.body.dataset.view !== "run") return; const key = target.dataset.toggle === "gantt" ? "ganttCollapsed" : target.dataset.toggle === "agents" ? "agentsCollapsed" : target.dataset.toggle === "logs" ? "logsCollapsed" : ""; if (!key) return; state.runLayout[key] = !state.runLayout[key]; saveRunLayout(); renderRun(); return; } if (target.dataset.act) { if (target.dataset.act === "steer") { const input = $("subagent-steer"); const message = input?.value.trim(); if (!message) return; sendAction("steer", { payload: { message } }); if (input) input.value = ""; return; } const name = target.closest("[data-name]")?.dataset.name; sendAction(target.dataset.act, name ? { name } : {}); return; } if (target.dataset.agentDetails !== undefined) { state.inspMode = "agent"; state.inspSig = null; renderAgent(); return; } if (target.dataset.script) { openScript(); return; } if (target.dataset.pane) { const found = selected(); const agent = found ? found.target.kind === "subagent" ? subagentAgent(found.record) : found.record.run.agents.find((candidate) => candidate.id === state.currentAgent) : undefined; const entries = found && agent ? agentEntries(found.record, agent) : []; if (target.closest("#tool-tabs")) { state.toolPane = target.dataset.pane; state.inspMode = "event"; renderInspector(entries[state.selectedEvent || 0] ? entryDetails(entries[state.selectedEvent || 0], agent, entries) : undefined, agent, entries); } else { state.sysPane = target.dataset.pane; state.inspMode = "agent"; if (agent) renderSystemPane(agent); } return; } if (target.dataset.run) { selectRun(target.dataset.run); return; } if (target.dataset.subagent) { selectSubagent(target.dataset.subagent); return; } if (target.dataset.agent && document.body.dataset.view === "run") { state.currentAgent = target.dataset.agent; state.selectedEvent = null; state.inspMode = "event"; setView("agent"); } });
+    document.addEventListener("toggle", (event) => { const target = event.target; if (!(target instanceof HTMLDetailsElement)) return; const key = target.dataset.workflowArgs ?? target.dataset.call; const open = target.dataset.workflowArgs !== undefined ? state.workflowArgumentsOpen : target.dataset.call !== undefined ? state.callsOpen : undefined; if (!open) return; if (target.open) open.add(key); else open.delete(key); }, true);
+    document.querySelector(".app").addEventListener("click", (event) => { const target = event.target.closest("[data-sidebar-group], [data-sidebar-section], [data-copy-share], [data-act], [data-run], [data-subagent], [data-agent], [data-agent-details], [data-script], [data-pane], [data-toggle]"); if (!target) return; if (target.dataset.copyShare) { const input = target.previousElementSibling; navigator.clipboard?.writeText(target.dataset.copyShare).then(() => showToast("copied"), () => { input?.select(); showToast("press Ctrl+C to copy"); }); return; } if (target.dataset.sidebarGroup !== undefined || target.dataset.sidebarSection !== undefined) { const key = target.dataset.sidebarGroup ?? target.dataset.sidebarSection; if (state.sidebarCollapsed.has(key)) state.sidebarCollapsed.delete(key); else state.sidebarCollapsed.add(key); saveSidebarCollapsed(); renderSidebar(); return; } if (target.dataset.toggle) { if (document.body.dataset.view !== "run") return; const key = target.dataset.toggle === "gantt" ? "ganttCollapsed" : target.dataset.toggle === "agents" ? "agentsCollapsed" : target.dataset.toggle === "calls" ? "callsCollapsed" : target.dataset.toggle === "logs" ? "logsCollapsed" : ""; if (!key) return; state.runLayout[key] = !state.runLayout[key]; saveRunLayout(); renderRun(); return; } if (target.dataset.act) { if (target.dataset.act === "steer") { const input = $("subagent-steer"); const message = input?.value.trim(); if (!message) return; sendAction("steer", { payload: { message } }); if (input) input.value = ""; return; } const name = target.closest("[data-name]")?.dataset.name; sendAction(target.dataset.act, name ? { name } : {}); return; } if (target.dataset.agentDetails !== undefined) { state.inspMode = "agent"; state.inspSig = null; renderAgent(); return; } if (target.dataset.script) { openScript(); return; } if (target.dataset.pane) { const found = selected(); const agent = found ? found.target.kind === "subagent" ? subagentAgent(found.record) : found.record.run.agents.find((candidate) => candidate.id === state.currentAgent) : undefined; const entries = found && agent ? agentEntries(found.record, agent) : []; if (target.closest("#tool-tabs")) { state.toolPane = target.dataset.pane; state.inspMode = "event"; renderInspector(entries[state.selectedEvent || 0] ? entryDetails(entries[state.selectedEvent || 0], agent, entries) : undefined, agent, entries); } else { state.sysPane = target.dataset.pane; state.inspMode = "agent"; if (agent) renderSystemPane(agent); } return; } if (target.dataset.run) { selectRun(target.dataset.run); return; } if (target.dataset.subagent) { selectSubagent(target.dataset.subagent); return; } if (target.dataset.agent && document.body.dataset.view === "run") { state.currentAgent = target.dataset.agent; state.selectedEvent = null; state.inspMode = "event"; setView("agent"); } });
     document.addEventListener("visibilitychange", () => { sendFocus(); if (!document.hidden) scheduleRender(); });
     // Phones show one sheet at a time, so picking something moves to the sheet that shows its result.
     const phone = matchMedia("(max-width: 900px)");

package/dist/trajectory/src/index.js

@@ -1,15 +1,16 @@
 import { spawn } from "node:child_process";
 import { createHash } from "node:crypto";
-import { mkdir, open, readFile, rm } from "node:fs/promises";
+import { mkdir, readFile, rm } from "node:fs/promises";
 import { existsSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
 import { getAgentDir } from "@earendil-works/pi-coding-agent";
 import { clearTrajectoryHost, setTrajectoryHost } from "../../src/trajectory-host-handle.js";
-import { processAlive } from "../../src/session-lease.js";
+import { atomicJson } from "../../src/io.js";
 import { errorText, isNodeError, object, positiveInteger } from "../../src/utils.js";
-import { isTrajectoryAction, isTrajectoryTarget, trajectoryActionError, TRAJECTORY_MAX_TRANSCRIPT_BYTES } from "../../src/trajectory.js";
+import { isTimingTranscriptEntry, isTrajectoryAction, isTrajectoryTarget, trajectoryActionError, TRAJECTORY_MAX_TRANSCRIPT_BYTES } from "../../src/trajectory.js";
 import { shareTrajectoryRun } from "./export.js";
+import { identityOf, processConfirmed, processLive, processStart, recordStartupServer, releaseStartupMutex, tryAcquireStartupMutex } from "./startup-mutex.js";
 const DEFAULT_TRAJECTORY_PORT = 7432;
 const TRAJECTORY_IDLE_EXIT_MS = 5 * 60 * 1000;
 const TRAJECTORY_LOCK_NAME = "trajectory.lock";
@@ -54,127 +55,166 @@ function signalProcess(pid, signal) {
             throw error;
     }
 }
-async function stopStaleServer(lock) {
-    // During startup, the lock can name the current Pi process rather than the detached server.
-    if (lock.pid === process.pid)
+// Called right after /health named this pid; its start, read now, tells a later reuse of the pid apart before SIGKILL.
+// Returns only once the server is gone: a server that may still serve keeps its lock and port, so nothing replaces it.
+async function stopStaleServer(pid, port) {
+    const server = identityOf(pid);
+    if (!processLive(server))
         return;
-    // NOTE: after a reboot the pid can belong to an unrelated process; startedAt (checked against /proc ctime) is the only proof it is still ours.
-    if (!await processAlive(lock.pid, lock.startedAt))
-        return;
-    signalProcess(lock.pid, "SIGTERM");
+    signalProcess(pid, "SIGTERM");
     const deadline = Date.now() + 3000;
     while (Date.now() < deadline) {
-        if (!await serverHealthy(lock.port))
+        if (!await serverHealthy(port))
             break;
         const remaining = deadline - Date.now();
         if (remaining <= 0)
             break;
         await delay(Math.min(50, remaining));
     }
-    if (await processAlive(lock.pid, lock.startedAt))
-        signalProcess(lock.pid, "SIGKILL");
+    // Without a confirmed identity the pid may now be an unrelated process, so SIGTERM is never escalated.
+    if (processConfirmed(server))
+        signalProcess(pid, "SIGKILL");
+    for (let attempt = 0; attempt < 50 && processLive(server); attempt += 1)
+        await delay(20);
+    if (processLive(server))
+        throw new Error(`Stale Trajectory server ${String(pid)} on port ${String(port)} did not stop; stop it and retry`);
 }
+// Only a missing or malformed lock reads as absent; any other read error leaves its owner unknown, so it is thrown.
 async function readLock(path) {
+    let text;
     try {
-        const parsed = JSON.parse(await readFile(path, "utf8"));
-        if (!object(parsed) || !positiveInteger(parsed.pid) || !positiveInteger(parsed.port) || parsed.port > 65535)
-            return undefined;
-        const fingerprint = typeof parsed.fingerprint === "string" ? parsed.fingerprint : undefined;
-        const startedAt = positiveInteger(parsed.startedAt) ? parsed.startedAt : undefined;
-        return { pid: parsed.pid, port: parsed.port, ...(fingerprint === undefined ? {} : { fingerprint }), ...(startedAt === undefined ? {} : { startedAt }) };
+        text = await readFile(path, "utf8");
     }
     catch (error) {
         if (isNodeError(error, "ENOENT"))
             return undefined;
+        throw error;
+    }
+    let parsed;
+    try {
+        parsed = JSON.parse(text);
+    }
+    catch {
         return undefined;
     }
+    if (!object(parsed) || !positiveInteger(parsed.pid) || !positiveInteger(parsed.port) || parsed.port > 65535)
+        return undefined;
+    const fingerprint = typeof parsed.fingerprint === "string" ? parsed.fingerprint : undefined;
+    const startedAt = positiveInteger(parsed.startedAt) ? parsed.startedAt : undefined;
+    const start = typeof parsed.start === "string" ? parsed.start : undefined;
+    return { pid: parsed.pid, port: parsed.port, ...(fingerprint === undefined ? {} : { fingerprint }), ...(startedAt === undefined ? {} : { startedAt }), ...(start === undefined ? {} : { start }) };
+}
+// One holder runs at most two bounded health waits (60 × (300 + 50) ms each) and one stop (3 s, its final health check, and 1 s for its exit).
+const STARTUP_MUTEX_WAIT_MS = 2 * 60 * (300 + 50) + 3_000 + 1_000 + 1_000;
+async function acquireStartupMutex(lockPath) {
+    const deadline = Date.now() + STARTUP_MUTEX_WAIT_MS;
+    for (;;) {
+        const token = await tryAcquireStartupMutex(lockPath);
+        if (token !== undefined)
+            return token;
+        if (Date.now() >= deadline)
+            throw new Error("Another Pi is still starting Trajectory; retry");
+        // Contenders that saw each other both back off; a random delay keeps them from retrying in step.
+        await delay(25 + Math.floor(Math.random() * 50));
+    }
+}
+// Only a Trajectory server reports a fingerprint; any other service answering /health is never signalled.
+function isTrajectoryServer(health) { return health?.pid !== undefined && health.fingerprint !== undefined; }
+// A lock proves which process serves its port only when that process reports the same identity.
+function ownsLock(health, lock) {
+    return isTrajectoryServer(health) && health.pid === lock.pid && (lock.startedAt === undefined || lock.startedAt === health.startedAt);
 }
-// Only a server reporting this fingerprint counts: an older one still bound to the port would otherwise be adopted forever.
-async function waitForServer(port, fingerprint) {
-    let foreign = false;
-    for (let attempt = 0; attempt < 60; attempt += 1) {
+async function awaitHealth(port, accept, stopped = () => false) {
+    for (let attempt = 0; attempt < 60 && !stopped(); attempt += 1) {
         const health = await serverHealth(port);
-        if (health?.fingerprint === fingerprint)
-            return;
-        foreign = health !== undefined;
+        if (health !== undefined && accept(health))
+            return health;
         await delay(50);
     }
-    throw new Error(foreign ? `Trajectory port ${String(port)} is held by another Trajectory server; stop it and retry` : `Trajectory server did not start on port ${String(port)}`);
+    return undefined;
 }
-async function resolveExistingServer(lockPath, existing, fingerprint) {
-    const health = await serverHealth(existing.port);
-    if (health) {
-        if (existing.fingerprint === fingerprint && health.fingerprint === fingerprint)
-            return existing;
-        await stopStaleServer(existing);
-        await rm(lockPath, { force: true });
-        return undefined;
+async function startServer(lockPath, serverPath, fingerprint, port, startupToken) {
+    // NOTE: under a Bun-compiled pi binary process.execPath is the pi CLI, and Bun's node:http never writes the WebSocket 101 upgrade (oven-sh/bun#28157), so the server must run on a real node from PATH.
+    const child = spawn(process.versions.bun ? "node" : process.execPath, [serverPath, "--port", String(port), "--lock", lockPath, "--fingerprint", fingerprint, "--startup", startupToken], { detached: true, stdio: "ignore" });
+    const startup = { ended: false };
+    const exited = new Promise((resolve) => {
+        child.once("error", (error) => { startup.error = error; startup.ended = true; resolve(); });
+        child.once("exit", () => { startup.ended = true; resolve(); });
+    });
+    try {
+        if (child.pid !== undefined)
+            recordStartupServer(lockPath, startupToken, child.pid);
+        const health = await awaitHealth(port, (candidate) => candidate.pid === child.pid && candidate.fingerprint === fingerprint, () => startup.ended);
+        if (startup.error !== undefined)
+            throw startup.error;
+        // The server publishes its lock before it answers, so readiness is reported only once both agree.
+        const lock = health === undefined ? undefined : await readLock(lockPath);
+        // The detached server outlives this Pi only once it is ready; until then this Pi keeps waiting for its exit.
+        if (lock !== undefined && ownsLock(health, lock) && lock.port === port && lock.fingerprint === fingerprint) {
+            child.unref();
+            return { port };
+        }
+        throw new Error(`Trajectory server did not start on port ${String(port)}`);
     }
-    if (await processAlive(existing.pid, existing.startedAt)) {
-        try {
-            await waitForServer(existing.port, fingerprint);
-            return existing;
+    catch (error) {
+        // A server outliving its startup could later publish over another owner's lock, so it is stopped before the mutex is released.
+        if (!startup.ended) {
+            child.kill("SIGKILL");
+            await exited;
         }
-        catch {
-            // A live lock can still name the attaching process during startup; after the bounded wait, replace the unrecoverable startup owner and retry normally.
-            await stopStaleServer(existing);
+        if ((await readLock(lockPath))?.pid === child.pid)
             await rm(lockPath, { force: true });
-            return undefined;
+        throw error;
+    }
+}
+// Runs under the startup mutex, so no other Pi or idle server changes the lock meanwhile.
+async function startOrReuseServer(lockPath, serverPath, fingerprint, configuredPort, startupToken) {
+    const existing = await readLock(lockPath);
+    if (existing) {
+        let health = await serverHealth(existing.port);
+        // A live lock owner can be a server still starting or one too loaded to answer in time.
+        if (health === undefined && processLive(existing))
+            health = await awaitHealth(existing.port, (candidate) => candidate.pid === existing.pid);
+        const lock = health === undefined ? existing : await readLock(lockPath) ?? existing;
+        if (ownsLock(health, lock)) {
+            if (lock.fingerprint === fingerprint && health.fingerprint === fingerprint)
+                return { port: lock.port };
+            await stopStaleServer(health.pid, lock.port);
+        }
+        else if (!isTrajectoryServer(health) && lock.pid !== process.pid && processLive(lock)) {
+            // Unidentified is not dead: a suspended or overloaded owner can resume serving, so its lock stays and no second server is started beside it.
+            // Only a lock naming this Pi, left by an older release's startup reservation, is known not to name a server.
+            throw new Error(health === undefined ? `Trajectory server ${String(lock.pid)} named by ${lockPath} is alive but not answering on port ${String(lock.port)}; retry once it answers, or remove the lock if that process is not a Trajectory server` : `Trajectory port ${String(lock.port)} is held by another Trajectory server; stop it and retry`);
         }
+        // Any other owner is dead, or another Trajectory server holds its port, so it is never signalled.
+        await rm(lockPath, { force: true });
     }
-    await rm(lockPath, { force: true });
-    return undefined;
+    const occupant = await serverHealth(configuredPort);
+    if (occupant !== undefined) {
+        if (!isTrajectoryServer(occupant))
+            throw new Error(`Trajectory port ${String(configuredPort)} is held by another Trajectory server; stop it and retry`);
+        if (occupant.fingerprint === fingerprint && occupant.startedAt !== undefined) {
+            // This server lost its lock; restoring it keeps every healthy server named by the lock.
+            //NOTE: its start is read just after /health named the pid; a reuse of that pid in between is not excluded.
+            await atomicJson(lockPath, { pid: occupant.pid, port: configuredPort, fingerprint, startedAt: occupant.startedAt, start: processStart(occupant.pid) });
+            return { port: configuredPort };
+        }
+        await stopStaleServer(occupant.pid, configuredPort);
+    }
+    return startServer(lockPath, serverPath, fingerprint, configuredPort, startupToken);
 }
+//NOTE: Pi releases without the startup mutex do not take it, so a concurrent start by one of them is not excluded.
 async function ensureTrajectoryServer(agentDir, configuredPort) {
     const lockPath = trajectoryLockPath(agentDir);
     const serverPath = trajectoryServerPath();
     const fingerprint = await trajectoryFingerprint(serverPath);
     await mkdir(dirname(lockPath), { recursive: true, mode: 0o700 });
-    const existing = await readLock(lockPath);
-    if (existing) {
-        const reused = await resolveExistingServer(lockPath, existing, fingerprint);
-        if (reused)
-            return reused;
-    }
-    let lockHandle;
+    const startupToken = await acquireStartupMutex(lockPath);
     try {
-        lockHandle = await open(lockPath, "wx", 0o600);
-        await lockHandle.writeFile(`${JSON.stringify({ pid: process.pid, port: configuredPort, fingerprint, startedAt: Date.now() })}\n`, "utf8");
-    }
-    catch (error) {
-        if (isNodeError(error, "EEXIST")) {
-            const raced = await readLock(lockPath);
-            if (raced) {
-                const reused = await resolveExistingServer(lockPath, raced, fingerprint);
-                if (reused)
-                    return reused;
-            }
-            await rm(lockPath, { force: true });
-            lockHandle = await open(lockPath, "wx", 0o600);
-            await lockHandle.writeFile(`${JSON.stringify({ pid: process.pid, port: configuredPort, fingerprint, startedAt: Date.now() })}\n`, "utf8");
-        }
-        else {
-            throw error;
-        }
+        return await startOrReuseServer(lockPath, serverPath, fingerprint, configuredPort, startupToken);
     }
     finally {
-        await lockHandle?.close();
-    }
-    // A server can answer on the port without owning the lock, e.g. one whose lock was lost; it reports its pid, so replace it before spawning.
-    const occupant = await serverHealth(configuredPort);
-    if (occupant?.pid !== undefined && occupant.fingerprint !== fingerprint)
-        await stopStaleServer({ pid: occupant.pid, port: configuredPort, ...(occupant.startedAt === undefined ? {} : { startedAt: occupant.startedAt }) });
-    try {
-        // NOTE: under a Bun-compiled pi binary process.execPath is the pi CLI, and Bun's node:http never writes the WebSocket 101 upgrade (oven-sh/bun#28157), so the server must run on a real node from PATH.
-        const child = spawn(process.versions.bun ? "node" : process.execPath, [serverPath, "--port", String(configuredPort), "--lock", lockPath, "--fingerprint", fingerprint], { detached: true, stdio: "ignore" });
-        const startupError = new Promise((_resolve, reject) => { child.once("error", reject); });
-        child.unref();
-        await Promise.race([waitForServer(configuredPort, fingerprint), startupError]);
-        return { port: configuredPort };
-    }
-    catch (error) {
-        await rm(lockPath, { force: true });
-        throw error;
+        await releaseStartupMutex(lockPath, startupToken);
     }
 }
 function trajectoryWebSocket() {
@@ -202,7 +242,6 @@ const ACTIVE_POLL_MS = 1_000;
 const IDLE_POLL_MS = 10_000;
 const RECONNECT_INITIAL_DELAY_MS = 100;
 const RECONNECT_MAX_DELAY_MS = 5_000;
-function isTimingEntry(value) { return object(value) && value.type === "custom" && value.customType === "pi-workflows:tool-timing"; }
 function boundedString(value, maxLength = 1024) {
     if (typeof value !== "string")
         return value;
@@ -213,7 +252,7 @@ const MAX_LIVE_STRING_BYTES = 64 * 1024;
 const MAX_LIVE_ARRAY_ENTRIES = 256;
 const MAX_LIVE_OBJECT_KEYS = 64;
 const LIVE_METADATA_ARRAY_KEYS = new Set(["agents", "runs", "subagents"]);
-const LIVE_TRUNCATABLE_ARRAY_KEYS = new Set(["events", "phaseHistory"]);
+const LIVE_TRUNCATABLE_ARRAY_KEYS = new Set(["events", "phaseHistory", "scriptCalls"]);
 const LIVE_METADATA_OBJECT_KEYS = new Set(["transcripts"]);
 function liveValueWillBeBounded(value, key = "", depth = 0) {
     if (typeof value === "string")
@@ -238,7 +277,7 @@ function liveValueWillBeBounded(value, key = "", depth = 0) {
 const MAX_LIVE_TIMING_BYTES = 64 * 1024;
 /** Keeps the newest timing entries: dropping the oldest leaves a live agent's gantt growing instead of frozen. */
 function boundedTiming(value) {
-    const entries = Array.isArray(value) ? value.filter(isTimingEntry) : [];
+    const entries = Array.isArray(value) ? value.filter(isTimingTranscriptEntry) : [];
     const retained = [];
     let bytes = 2;
     for (let index = entries.length - 1; index >= 0; index -= 1) {
@@ -513,23 +552,14 @@ export function createTrajectoryController(agentDir) {
         }
     };
     const transcriptFromFallback = async (input, request) => {
-        const boundedTranscriptResult = (result) => {
-            try {
-                if (result.status === "available" && Buffer.byteLength(JSON.stringify(result.entries)) > TRAJECTORY_MAX_TRANSCRIPT_BYTES)
-                    return { status: "oversized", revision: result.revision, entries: [], error: "Transcript is too large" };
-            }
-            catch {
-                return { status: "failed", revision: result.revision, entries: [], error: "Transcript failed" };
-            }
-            return result;
-        };
-        const boundedResult = (entries, revision) => { try {
-            if (Buffer.byteLength(JSON.stringify(entries)) > TRAJECTORY_MAX_TRANSCRIPT_BYTES)
-                return { status: "oversized", revision, entries: [], error: "Transcript is too large" };
+        const oversized = (entries, revision) => { try {
+            return Buffer.byteLength(JSON.stringify(entries)) > TRAJECTORY_MAX_TRANSCRIPT_BYTES ? { status: "oversized", revision, entries: [], error: "Transcript is too large" } : undefined;
         }
         catch {
             return { status: "failed", revision, entries: [], error: "Transcript failed" };
-        } return { status: entries.length ? "available" : "empty", revision, entries }; };
+        } };
+        const boundedTranscriptResult = (result) => (result.status === "available" ? oversized(result.entries, result.revision) : undefined) ?? result;
+        const boundedResult = (entries, revision) => oversized(entries, revision) ?? { status: entries.length ? "available" : "empty", revision, entries };
         if (input.loadTranscript) {
             const result = await input.loadTranscript(request);
             const revisionChanged = request.revision !== undefined && result.revision !== request.revision && result.status !== "missing" && result.status !== "failed" && result.status !== "oversized" && result.status !== "disconnected";