toolroll 0.9.91 → 0.9.93

package/dist/surface.js

@@ -116,12 +116,12 @@ const GUIDE_ROWS = [
         notableReasons: ["usage", "ambiguous", "drift", "timeout", "check-failed", "ci-failed", "tree-mismatch", "conflict", "mixed-runtime", "gate-held", "after-swap"] },
     operator("demo", "a seeded throwaway sandbox"),
     ...KNOWLEDGE_DESCRIPTORS.map(spec => ({ invocation: `knowledge ${spec.action}`, synopsis: spec.synopsis, audience: "audience" in spec ? spec.audience : "agent", agentMayInvoke: !("audience" in spec), mutation: spec.mutation, flags: spec.flags, ...(spec.takesQuery ? { positionals: [{ name: "query", required: true, meaning: "search text or source file for impact" }] } : {}) })),
-    ...MEMORY_DESCRIPTORS.map(spec => ({ invocation: `memory ${spec.action}`, synopsis: spec.synopsis, audience: "agent", agentMayInvoke: true, mutation: spec.mutation, flags: spec.flags, ...(spec.takesQuery ? { positionals: [{ name: "query", required: true, meaning: "search text, a decision id, or the decision sentence" }] } : {}) })),
+    ...MEMORY_DESCRIPTORS.map(spec => ({ invocation: `memory ${spec.action}`, synopsis: spec.synopsis, audience: "audience" in spec ? spec.audience : "agent", agentMayInvoke: !("audience" in spec), mutation: spec.mutation, flags: spec.flags, ...(spec.takesQuery ? { positionals: [{ name: "query", required: true, meaning: "search text, a decision id, or the decision sentence" }] } : {}) })),
     // Flow writes carry an approver's credential: an agent runs them only as the person asked, after showing the preview.
     // Letting a flow run on its own raises authority: the owner's own password ceremony, never an agent's act.
     ...FLOWS_DESCRIPTORS.map(spec => ({ invocation: `flows ${spec.action}`, synopsis: spec.synopsis, audience: "audience" in spec ? spec.audience : "agent", agentMayInvoke: !("audience" in spec), mutation: spec.mutation, flags: spec.flags,
         ...("positionals" in spec ? { positionals: spec.positionals.map(name => ({ name, required: true, meaning: name === "flow" ? "a flow id from flows list" : name === "card" ? "a card id from flows show" : name === "on|off" ? "on signs the grant, off revokes it" : "a trigger id from flows show, or for trigger add its settings as JSON, a JSON file, or -" })) } : {}) })),
-    ...MODELS_DESCRIPTORS.map(spec => ({ invocation: `models ${spec.action}`, synopsis: spec.synopsis, audience: "agent", agentMayInvoke: true, mutation: spec.mutation, flags: spec.flags, ...(spec.takesQuery ? { positionals: [{ name: "target", required: true, meaning: "the CLI to update, or on/off" }] } : {}) })),
+    ...MODELS_DESCRIPTORS.map(spec => ({ invocation: `models ${spec.action}`, synopsis: spec.synopsis, audience: "agent", agentMayInvoke: true, mutation: spec.mutation, flags: spec.flags, ...(spec.takesQuery ? { positionals: [{ name: "target", required: true, meaning: "positional" in spec ? spec.positional : "the CLI to update, or on/off" }] } : {}) })),
     // ---- the queue (agent surface) ----
     { invocation: "status", synopsis: "running work, queued reasons, results to review, the latest release check and plan windows", audience: "agent", agentMayInvoke: true, mutation: "none", flags: [jsonFlag, dbFlag] },
     { invocation: 'health', synopsis: 'recent server latency, event-loop delay, SQLite writes and live streams from the saved connection; instance operators only', audience: 'agent', agentMayInvoke: true, mutation: 'none', flags: [jsonFlag, { name: 'profile', takesValue: true, meaning: 'saved server connection' }] },
@@ -315,7 +315,7 @@ const GUIDE_ROWS = [
             { name: "days", takesValue: true, meaning: "also finished tasks untouched this many days (1–365, default 14)" },
             { name: "except", takesValue: true, meaning: "comma-separated tasks from the list to leave out" }],
         notableReasons: ["usage", "not-an-approver", "empty"] },
-    operator("run settle", "last resort for a finished run whose process witness can't be proven either way: records your --why in the ledger; refuses while any of the run's processes is alive"),
+    operator("run settle", "have Toolroll check again whether a finished run stopped: it looks at this computer's processes itself, stops only leftovers proven the run's, and settles the run when none is left (nobody vouches; --why is ignored)"),
     operator("task resume", "resume the exact stopped attempt (--run <id>): refuses until it is quiescent, lifts only that stop's hold, approves nothing — the next pass re-proves the scope and inherits the draft with fresh proof"),
     // ---- leases (runner surface) ----
     { invocation: "claim", synopsis: "take one ready, approved task; answers with a lease", audience: "agent", agentMayInvoke: true, mutation: "keyed",
@@ -460,6 +460,11 @@ const REMOTE_NO = [
     "serve", "serve check-public", "watch", "up", "onboard", "daemon", "background on", "background off", "background show", "stop", "open", "welcome", "password set", "doctor", "bridge", "setup show", "setup clear",
     // Provider keys and model tools on this machine.
     "providers", "keys status", "keys set", "keys clear", "keys verify", "keys auth", "models check", "models update", "models watch",
+    // Result records read with this machine's sign-in: project access, never a token's narrower scope.
+    "models results",
+    // Earned rules (results routing): the project's own results, and its rules' doors, with this machine's sign-in; a
+    // remote token never changes which models run (the rule door refuses it in the write too).
+    "models suggestions", "models accept", "models dismiss", "models turn-off", "models keep", "models try", "models stop",
     // Publication and release.
     "publish", "publish status", "publish merge", "publish refire", "publish unblock", "publish rearm", "publish grant", "publish revoke",
     // A repository's authority grants on this machine.
@@ -474,7 +479,9 @@ const REMOTE_NO = [
 const REMOTE_STEP_UP = [
     "proposals", "decide", "task regate", "task approve", "task accept", "task merge", "run settle",
     // Putting tasks away and restoring them: a person's act, in the console or on the server's machine.
-    "task archive", "task restore", "task tidy", "knowledge apply", "memory apply",
+    "task archive", "task restore", "task tidy", "knowledge apply",
+    // Changing project memory: a person's own password, never a token (the lead only proposes it in chat).
+    "memory decide", "memory retire", "memory apply",
     "flows card approve", "flows card send-back", "flows script approve", "flows unattended",
     "approver list", "approver add", "people list", "people invite", "people projects", "people revoke",
     // A person's own API tokens: a password step-up on the server's machine, or the console. Never with a token.
@@ -491,7 +498,7 @@ const REMOTE_YES = [
     "config show", "verify show", "intake show", "intake run", "intake preview", "intake pr-comments", "template list", "template show",
     "webhook status", "review show", "chat-approval show", "chat-approval off",
     "knowledge search", "knowledge impact", "knowledge refresh",
-    "memory search", "memory decisions", "memory show", "memory decide", "memory retire", "memory review", "memory status",
+    "memory search", "memory decisions", "memory show", "memory review", "memory status",
     "flows list", "flows show", "flows export", "flows create", "flows edit", "flows trigger add", "flows trigger pause", "flows trigger resume",
     "flows trigger remove", "flows card add", "flows archive",
     "models status", "models list",
@@ -530,6 +537,11 @@ const LOCAL_ALLOW = {
     "task archive": "archive-task",
     "task restore": "archive-task",
     "repos add": "add-project",
+    // Results routing: lowering only (route-store.ts proveRuleDoor checks the act is exactly this one).
+    "models suggestions": "read",
+    "models turn-off": "routing-turn-off",
+    "models dismiss": "dismiss-suggestion",
+    "models stop": "trial-stop",
 };
 /** Every guide row's local-owner mark, by invocation: `allow` with its act, else `deny`. Exported for the inventory test. */
 export const LOCAL_POLICY = new Map(GUIDE_ROWS.map(row => [row.invocation,

package/dist/task-outcome-cli.js

@@ -4,11 +4,13 @@ import { assignmentOf, checkAssignmentAsOperator } from './assignment.js';
 import { assignmentForCoordinator, coordinatorToken } from './assignment-adapters.js';
 import { authenticateCoordinator } from './coordinator.js';
 import { reproveApprover } from './principal.js';
-import { authenticateApprover, personProof } from './scope.js';
+import { authenticateApprover, machineActing, personProof } from './scope.js';
+import { currentActor } from './actor.js';
 import { requestResultChanges } from './result-actions.js';
 import { commentSourceKey, revisionSourceOf } from './result-review.js';
 import { envelopeJson } from './envelope.js';
 import { completeAndOpenPullRequest, mergeAsPerson, MERGE_NEEDS_PERSON, newestPullRequestOf, publishingOf, pullRequestViewOf } from './pull-request-flow.js';
+import { proveRunStopped } from './stop-proof.js';
 export async function runTaskOutcomeCommand(action, positional, flags, context) {
     const command = `task ${action}`;
     const emit = (result, lines) => {
@@ -35,6 +37,10 @@ export async function runTaskOutcomeCommand(action, positional, flags, context)
     if (credential.token === null && operator === null)
         return fail('unauthenticated', 'Sign in with up, or supply --as and --token. Agents can use --token-file or --token-env with a scoped credential.');
     const { store, now, evidenceRoot } = context;
+    // A result whose build Toolroll hasn't yet proven stopped: it looks now (stop-proof.ts), so completing never waits
+    // on a person to confirm it, only on Toolroll's own check (a refusal says "stopping" until it lands).
+    if (action === 'complete')
+        await proveTaskBuildsStopped(store, task, () => now);
     return store.transact(() => {
         const coordinator = credential.token === null ? null : authenticateCoordinator(store, credential.token);
         if (coordinator !== null && !coordinator.ok)
@@ -110,7 +116,9 @@ export async function runTaskOutcomeCommand(action, positional, flags, context)
         // The shared action returns exact replays before stale-family checks. An
         // older result cannot create another child once a newer revision is current.
         const revised = requestResultChanges(store, evidenceRoot, { run: run.id, source, actor, repos,
-            batch: '', note: feedback, path: '', line: '', request: key, allowMode: operator !== null }, now);
+            batch: '', note: feedback, path: '', line: '', request: key, allowMode: operator !== null,
+            // A verified approver's own words; a scoped credential or the lead's token is the lead; a remote API run a machine.
+            origin: coordinator !== null || currentActor()?.lead === true ? 'lead' : machineActing() ? 'machine' : 'person' }, now);
         if (!revised.ok)
             return fail(revised.status === 400 ? 'usage' : 'stale-result', revised.message);
         const scope = store.getScope(revised.id);
@@ -169,3 +177,17 @@ export function pullRequestLines(view) {
         return [];
     return [`  pull request: ${view.prUrl ?? 'opening'} · ${view.label}`, `    ${view.detail}`, ...(view.mergeCommit === null ? [] : [`    merge commit: ${view.mergeCommit}`])];
 }
+/** Toolroll's own stop proof for every finished build of a task's family whose stop it can prove here. Best effort: a
+ * failure leaves the completion's own refusal to say what is still unproven. */
+async function proveTaskBuildsStopped(store, task, clock) {
+    try {
+        const family = store.taskFamilyOf(task, null, true);
+        const refs = family === null ? [store.lookupRef(task)?.id].filter((one) => one !== undefined) : family.versions.map(version => version.refId);
+        for (const ref of refs)
+            for (const run of store.runsFor(ref)) {
+                if (run.outcome !== null && store.stopQuiescenceFact(run.id)?.cause !== undefined)
+                    await proveRunStopped(store, run.id, clock);
+            }
+    }
+    catch { /* the completion below refuses with what is still unproven */ }
+}

package/dist/toolroll-update-ui.js

@@ -24,7 +24,7 @@ export function updateStepsHtml(j, running, options = {}) {
     const items = UPDATE_STEPS.map((step, i) => {
         const state = j.phase === "complete" || i < last ? "done" : i > last ? "waiting" : failed ? "failed" : finished ? "waiting" : "now";
         // Only where it adds something: which work it waits for, or what went wrong.
-        const detail = state === "now" && step === "draining" && j.waiting ? html `${j.waiting.on}.${j.waiting.action ? html ` If nothing of it is running, run <code class="k-mono">${j.waiting.action}</code>.` : ""}`
+        const detail = state === "now" && step === "draining" && j.waiting ? html `${j.waiting.on}.${j.waiting.action ? html ` To have Toolroll check again, run <code class="k-mono">${j.waiting.action}</code>.` : ""}`
             : (state === "now" && step === "draining") || (state === "failed" && j.phase === "rolling-back") ? html `${j.detail}` : "";
         return settingsRecord({ title: STEP_WORDS[step], sub: detail, when: STEP_CHIP[state], attrs: { "data-step": step, "data-state": state } });
     });
@@ -43,7 +43,7 @@ function outcomeHtml(j) {
     // Stopped on a finished run it can't show has ended: what is in the way and the one command, nothing more.
     if (j.phase === "refused" && j.waiting) {
         return settingsSection({ id: "update-outcome", title: `Update to ${j.to.version} stopped: run #${j.waiting.run} is in the way`,
-            body: html `<div data-update-outcome="stopped">${infoLine(`${j.waiting.on}. New work resumed.`)}${j.waiting.action ? note(html `If nothing of it is running, run <code class="k-mono">${j.waiting.action}</code>, then update again.`) : note("Update again once it has stopped.")}</div>` });
+            body: html `<div data-update-outcome="stopped">${infoLine(`${j.waiting.on}. New work resumed.`)}${j.waiting.action ? note(html `To have Toolroll check again, run <code class="k-mono">${j.waiting.action}</code>, then update again.`) : note("Update again once it has stopped.")}</div>` });
     }
     const title = j.phase === "refused" ? `Didn't update to ${j.to.version}` : j.phase === "restored" ? `Update to ${j.to.version} didn't finish` : "The update needs attention";
     return settingsSection({ id: "update-outcome", title, state: html ` ${chip("Failed", "failed")}`,

package/dist/toolroll-update.js

@@ -39,7 +39,7 @@ import { installLaunchdService, launchdPlist, stopLaunchdService, writeFileDurab
 import { LAPTOP_LABEL_DARWIN, LAPTOP_LABEL_LINUX, unitFlag } from "./laptop-service.js";
 import { processGone, readRunnerProcessRecord, realProcessProbe, recordStandsFor, RUNNER_PROCESS_DIR } from "./runner-process.js";
 import { NAME } from "./names.js";
-import { orphanKey, updateSafeSchema, V117_RENAMED_COLUMNS, V117_RENAMED_TABLES, V118_MERGED_TABLES, V119_ADDED_COLUMNS, V120_ADDED_COLUMNS, V121_ADDED_COLUMNS, V122_ADDED_COLUMNS, V123_ADDED_COLUMNS, V125_ADDED_COLUMNS, V126_ADDED_COLUMNS, V128_ADDED_COLUMNS, V129_ADDED_COLUMNS, V134_ADDED_COLUMNS, V135_ADDED_COLUMNS, V138_ADDED_COLUMNS, V143_ADDED_COLUMNS, V147_ADDED_COLUMNS } from "./store.js";
+import { orphanKey, updateSafeSchema, V117_RENAMED_COLUMNS, V117_RENAMED_TABLES, V118_MERGED_TABLES, V119_ADDED_COLUMNS, V120_ADDED_COLUMNS, V121_ADDED_COLUMNS, V122_ADDED_COLUMNS, V123_ADDED_COLUMNS, V125_ADDED_COLUMNS, V126_ADDED_COLUMNS, V128_ADDED_COLUMNS, V129_ADDED_COLUMNS, V134_ADDED_COLUMNS, V135_ADDED_COLUMNS, V138_ADDED_COLUMNS, V143_ADDED_COLUMNS, V147_ADDED_COLUMNS, V148_ADDED_COLUMNS } from "./store.js";
 import { isNewer, REGISTRY } from "./releases.js";
 import { markNeverIndex } from "./never-index.js";
 import { readRuntimeUpdateJournal, RUNTIME_PHASES, RUNTIME_UPDATE_STEPS, stagedStartedAt, updaterStartingOf } from "./contracts/update-journal.js";
@@ -413,7 +413,9 @@ const HISTORY_RULES = {
     // v128: a lead message's pending flag reads waiting exactly for a team queue's unadmitted operator message.
     // v147: an account's password flag reads set (1), a browser session reads full with no local binding or key, and the
     // console record reads no key epoch.
-    ...Object.fromEntries(Object.entries(addedColumns(V119_ADDED_COLUMNS, V120_ADDED_COLUMNS, V121_ADDED_COLUMNS, V122_ADDED_COLUMNS, V123_ADDED_COLUMNS, V125_ADDED_COLUMNS, V126_ADDED_COLUMNS, V128_ADDED_COLUMNS, V129_ADDED_COLUMNS, V134_ADDED_COLUMNS, V135_ADDED_COLUMNS, V138_ADDED_COLUMNS, V143_ADDED_COLUMNS, V147_ADDED_COLUMNS))
+    // v148: a run's execution class and a revision's origin read none (NULL), and a project's suggestion switch reads on.
+    // v149 adds tables only (the learning-loops records): no saved row changes, so no rule is needed for it.
+    ...Object.fromEntries(Object.entries(addedColumns(V119_ADDED_COLUMNS, V120_ADDED_COLUMNS, V121_ADDED_COLUMNS, V122_ADDED_COLUMNS, V123_ADDED_COLUMNS, V125_ADDED_COLUMNS, V126_ADDED_COLUMNS, V128_ADDED_COLUMNS, V129_ADDED_COLUMNS, V134_ADDED_COLUMNS, V135_ADDED_COLUMNS, V138_ADDED_COLUMNS, V143_ADDED_COLUMNS, V147_ADDED_COLUMNS, V148_ADDED_COLUMNS))
         .map(([table, rule]) => [table, { ...EARLIER_RULES[table], ...rule }])),
 };
 /** The tables a migration may remove whole (v115's removed features). */

package/dist/work-index.js

@@ -41,7 +41,7 @@ const AGENT_ASK = `(SELECT q.id FROM agent_question q JOIN run qr ON qr.id=q.run
 const GATE_PAUSE_ON_TASK = pauseOnTaskProviders('p', 'c.ref_id');
 /** Index codes that wait on a person, as their need (needs-you.ts); dispatch codes come from task-status's own table. */
 const INDEX_NEED = {
-    'earlier-active': 'earlier-version', 'process-needs-attention': 'confirm-stopped', 'invalid-scope': 'define-task', 'decision-queue': 'questions',
+    'earlier-active': 'earlier-version', 'process-needs-attention': 'stop-unproven', 'invalid-scope': 'define-task', 'decision-queue': 'questions',
 };
 /** Every index code that can wait on a person. */
 const PERSON_CODES = ['history-problem', 'earlier-active', 'ready-to-check', 'waiting-decision', 'decision-queue', 'result-needs-attention',
@@ -572,7 +572,7 @@ function itemOf(row, principal, probe, leadName, readiness, readable = false) {
     const code = String(row['code']), id = String(row['id']);
     const custody = code === 'process-needs-attention' || (code === 'result-needs-attention' && n(row, 'custody_unresolved') === 1);
     const process = custody && probe !== undefined ? processNeedOf(probe(id)) : null;
-    const waiting = process !== null && 'wait' in process;
+    const waiting = (process !== null && 'wait' in process) || (custody && probe === undefined);
     const need = n(row, 'needs') === 1 && !waiting, running = n(row, 'family_running') === 1;
     // The lead's claim (lead-voice.ts): the row still reads its own stage, worded "<name> is on it".
     const lead = n(row, 'lead_claim') === 1 ? 'on-it' : n(row, 'lead_claim') === 2 ? 'lapsed' : null;
@@ -647,14 +647,11 @@ function itemOf(row, principal, probe, leadName, readiness, readable = false) {
         actionCode = 'open-result';
         actionLabel = 'Open result';
     }
+    // Toolroll proves a build stopped itself; only a stop it can't prove from here offers "Check again", never a vouch.
     if (process !== null && 'need' in process) {
-        actionCode = 'confirm-stopped';
+        actionCode = 'recheck-stopped';
         actionLabel = NEEDS[process.need].action.label;
     }
-    else if (custody && probe === undefined) {
-        actionCode = 'confirm-stopped';
-        actionLabel = NEEDS['confirm-stopped'].action.label;
-    }
     if (code === 'signed-out') {
         actionCode = 'sign-in';
         actionLabel = NEEDS['sign-in'].action.label;
@@ -668,7 +665,7 @@ function itemOf(row, principal, probe, leadName, readiness, readable = false) {
         target: { taskId: successor ?? (code === 'earlier-active' ? s(row, 'earlier_id') ?? id : actionCode === 'answer-decision' ? s(row, 'question_task') ?? id : id),
             runId: actionCode === 'answer-decision' ? n(row, 'question_run') || null : actionCode === 'answer-agent' ? n(row, 'agent_ask_run') || null : actionCode === 'inspect-run' && code === 'running' ? n(row, 'live_run') || null : actionCode === 'open-result' || actionCode === 'inspect-run' ? n(row, 'result_id') || null
                 : actionCode === 'reconcile-run' ? n(row, 'unfinished') || null : ['resume-run', 'inspect-stop'].includes(actionCode) ? n(row, 'stop_run') || null
-                    : actionCode === 'confirm-stopped' ? process?.build ?? (n(row, 'result_id') || null) : null,
+                    : actionCode === 'recheck-stopped' ? process?.build ?? (n(row, 'result_id') || null) : null,
             decisionId: actionCode === 'answer-decision' ? n(row, 'question_id') || null : null },
         access: read ? 'read' : principal === 'operator' ? 'operator-control' : ['answer-decision', 'answer-agent', 'write-scope', 'unhold'].includes(actionCode) ? 'proposal-only' : 'operator-handoff',
         retry: actionCode === 'reconcile-run' ? 'reconcile-before-retry' : read ? 'read-again' : 'refresh-before-acting' };
@@ -676,7 +673,8 @@ function itemOf(row, principal, probe, leadName, readiness, readable = false) {
     const coded = stageOfCode(code, { needsPerson: personNeed, planning: s(row, 'live_role') === 'planner', operatorHold: s(row, 'hold_kind') === 'operator' });
     const reading = process !== null
         ? 'wait' in process ? { stage: 'waiting', wait: process.wait } : { stage: 'needs-you', need: process.need }
-        : custody ? { stage: 'needs-you', need: 'confirm-stopped' }
+        // Not read here (no process probe): Toolroll's own stop check settles it, so it waits on no one.
+        : custody ? { stage: 'waiting', wait: 'build-checking' }
             : coded.stage === 'needs-you' && (coded.need === undefined || coded.need === 'other') && INDEX_NEED[code] !== undefined ? { stage: 'needs-you', need: INDEX_NEED[code] } : coded;
     const checkStatus = s(row, 'check_status'), verdict = s(row, 'proof_verdict');
     const finished = reading.stage === 'finished' || reading.stage === 'complete';

package/dist/work-summary.d.ts

@@ -8,9 +8,9 @@ export type WorkTarget = {
     decisionId: number | null;
 };
 export type WorkAction = {
-    /** `confirm-stopped`: an approver confirms a finished build stopped (behind the password);
+    /** `recheck-stopped`: Toolroll looks again whether a finished build stopped (nobody vouches for it);
      * `sign-in`: the agent's sign-in, in Settings; `answer-agent`: Reply to a running agent's question on the Agents board. */
-    code: DispatchAction | "inspect-task" | "inspect-run" | "inspect-stop" | "inspect-decisions" | "reconcile-run" | "open-pr" | "confirm-stopped" | "sign-in" | "answer-agent";
+    code: DispatchAction | "inspect-task" | "inspect-run" | "inspect-stop" | "inspect-decisions" | "reconcile-run" | "open-pr" | "recheck-stopped" | "sign-in" | "answer-agent";
     label: string;
     target: WorkTarget;
     access: "read" | "operator-control" | "proposal-only" | "operator-handoff";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "toolroll",
-  "version": "0.9.91",
+  "version": "0.9.93",
   "description": "A control plane for unattended coding agents — queue tasks, walk away, come back to pull requests. Agents build in leased worktrees, never touch your default branch, and interrupt you only for decisions that need a human.",
   "directories": {
     "doc": "docs"