@kylecheng3146/agent-ops 0.6.0 → 0.7.0

package/dist/runtime/src/install/harness.js

@@ -305,6 +305,49 @@ function subagentInstruction(descriptor) {
             return "Delegate it with this harness's subagent or task tool when it has one; otherwise run the same bounded steps sequentially in this session.";
     }
 }
+/**
+ * The `run` profile's hand-off of feature-sized work to `agent-ops run`.
+ * agy keeps the subtask flow: `run` launches only Claude Code or Codex.
+ * AGENTS.md is shared by Codex and opencode, so it carries the Codex form.
+ */
+function autoRunInstruction(descriptor, context) {
+    const file = descriptor.control.instructionFile;
+    if (!context.capabilities.includes("auto-run") || file === "GEMINI.md") {
+        return [];
+    }
+    const host = file === "CLAUDE.md" ? "claude" : "codex";
+    const start = `agent-ops run --goal-file .agent-ops/state/run-goal.md --host ${host} --wait`;
+    return [
+        "With the `run` profile, a change that needs more than five acceptance",
+        "criteria goes to `agent-ops run` instead of being split here; five or",
+        "fewer stay in this session. From the main checkout, before any `task",
+        "create` (its worktree makes `run` refuse with `WORKTREE_NESTED`), write",
+        "the user's prompt verbatim, then your proposed acceptance criteria marked",
+        "as proposals, to `.agent-ops/state/run-goal.md` (gitignored), and start",
+        ...(host === "claude"
+            ? [`\`${start}\``, "as a background shell command."]
+            : [
+                "it as a background shell command through the trusted outer host",
+                "runner, like review: the outer request uses",
+                "`sandbox_permissions: \"require_escalated\"`, and the shell form is",
+                `\`env -u CODEX_SANDBOX_NETWORK_DISABLED ${start}\`.`
+            ]),
+        "When it returns awaiting input, relay the question to the user, run",
+        "`agent-ops run respond <id> --question-id <q> --answer <text>`, and poll",
+        "`agent-ops run status <id>` until it finishes or asks again; on",
+        "completion report the result. If the start is refused with",
+        "`RUN_TARGET_REQUIRED`, `RUN_BACKGROUND_UNSUPPORTED`, `RUN_TARGET_DIRTY` or",
+        "`WORKTREE_NESTED`, say so in one line naming the code and fall back to",
+        "the subtask flow (`task create --parent`, then batch or advance).",
+        "`RUN_REPO_UNTRUSTED` means stop and ask the user; never work",
+        "around trust. If the run ends without completing (blocked, budget",
+        "exhausted, stopped, failing review), report the `agent-ops run status",
+        "<id>` state and code, the last `agent-ops run logs <id>` events, and the",
+        "exact `agent-ops run resume <id>` and `agent-ops run stop <id>` commands,",
+        "then stop: never resume automatically or take over the run's work.",
+        ""
+    ];
+}
 export function managedRules(descriptor, context) {
     const lines = [
         "# Loop Engineering",
@@ -322,7 +365,7 @@ export function managedRules(descriptor, context) {
         lines.push("For every change:", "", "1. Define two to five mechanically verifiable acceptance criteria.", "2. Inspect the smallest relevant scope and preserve unrelated changes.", "3. Apply the smallest safe change.", "4. Run evidence-producing verification for every criterion.", "5. Obtain complete independent review before claiming completion, via", "   `agent-ops review --task <task-id> --yes` (or the CLI's equivalent invocation).", "   Never call a", "   review-target CLI (agy, codex, claude) directly — direct calls skip", "   the enforced read-only sandbox flags and can hang or fail on command", "   permission prompts.", "   Start this command through the trusted outer host runner as its first", "   and only invocation. On Codex, the outer tool request must use", "   `sandbox_permissions: \"require_escalated\"` so network and loopback", "   are available; in that same request, unset", "   `CODEX_SANDBOX_NETWORK_DISABLED` and set `AGENT_OPS_HOST` when three", "   targets are configured. The Codex shell form is", "   `env -u CODEX_SANDBOX_NETWORK_DISABLED AGENT_OPS_HOST=<current-host> agent-ops review --task <task-id> --yes`.", "   Keep it a plain command line: to keep the output add `--output <file>`", "   (the JSON envelope is written there and one line is printed) instead of", "   a shell redirect, which the host can only allow by matching the whole", "   command string, task id included.", "   Do not run it once in the restricted sandbox and retry; changing the", "   variable inside that sandbox is not elevation.", "   Review runs exactly two fresh sessions: with three configured targets,", "   set `AGENT_OPS_HOST` to the current CLI id so it is excluded; with two", "   targets both run in configured order; with one target it runs twice.", "   The first session is necessary; a PASS then requires an adversarial", "   re-check carrying the first report. A first FAIL or NOT_RUN stops.", "", "Treat `.agent-ops/config.json` as verifier authority. Discovery output is", "only a proposal until a user confirms it. Repository commands require an", "exact matching trust record. Confirmed project init/update grants it", "automatically when verification commands are configured.", "");
     }
     if (context.capabilities.includes("task")) {
-        lines.push("Split a change that exceeds five acceptance criteria into subtasks:", "`agent-ops task create --parent <task-id>` records one, and", "`agent-ops task status --parent <task-id>` lists them. Each subtask", "carries its own criteria and verification; final review covers all", "tasks together when their worktrees are integrated;", "completing one never completes its parent.", "", "To review every active subtask of a parent together, run", "`agent-ops batch --parent <task-id> --yes` instead of one review per", "subtask. It needs what review needs: start it through the trusted outer", "host runner as its first and only invocation, with the same Codex", "elevation, and keep `--output <file>` instead of a shell redirect. It", "verifies, one task at a time, each task that lacks fresh evidence, and", "reviews a few tasks at once. Do not commit or edit the worktree while it", "runs: a moved HEAD voids every verify and review in the batch.", "For subagents in separate worktrees, use `task advance` instead.", "", "Delegate a subtask whose scope spans unknown files, broad scans, or an", "independent workstream instead of doing it inline. Name the bounded", "scope, the artifact to return, and the acceptance criteria it serves.", subagentInstruction(descriptor), "Read every returned artifact back yourself: integration, verification,", "and review stay with the coordinator.", "", "One writer at a time. `agent-ops verify` and `agent-ops review`", "fingerprint the working tree and report UNKNOWN when anything else", "writes it during the run, so delegate edits to one subagent at a time,", "or give each writer its own Git worktree and merge before verifying.", "Read-only delegation has no such limit and may run in parallel.", "", "Parallel conversations in one repository each need their own worktree.", "When `.agent-ops/config.json` sets `worktree.mode` to `auto`, run", "`agent-ops task create` from the main checkout before your first edit:", "it creates this session's worktree and puts the task there (`agent-ops", "worktree add <name> --session <session-id>` does the same without a", "task). Work only inside the printed path", "(Claude Code: EnterWorktree with that path; otherwise use it as every", "command's working directory). Commit, then run verify and review for", "each task with the printed `--base`; do not run `task complete` in the", "worktree. Leave it (Claude Code: ExitWorktree with action keep) and run", "`agent-ops worktree finish <name>` from the main checkout: it completes", "every task in the worktree, subtasks first, then merges. A finish that", "reports an incomplete task names it; verify or review that task again", "in the worktree and finish again. A finish that reports a conflict is", "resolved once, as a new task whose criteria cover both intents; if its", "verification or review fails, reset as instructed and report.", "", "For a session with writing subagents, record each child task's complete", "modification intent with `task create --intent <text>` before editing.", "Each child commits and runs local `verify --task` in its own worktree.", "From the main checkout run `agent-ops task advance --task <parent-id>", "--session <session-id> --yes` through the trusted outer host runner,", "unsetting `CODEX_SANDBOX_NETWORK_DISABLED` on Codex. Advance integrates", "child commits and intent into the coordinator worktree, verifies every", "task on the final candidate, then runs one two-round tree review.", "Only `NOT_RUN / scope-too-large` falls back to per-task two-round review;", "a substantive FAIL must be fixed. After PASS, advance writes a local", "receipt under the main checkout's Git common directory and calls finish.", "If the target moves, it retries the final gate once. An unresolved", "failure preserves worktrees and cannot be reported as finished.", "");
+        lines.push("Split a change that exceeds five acceptance criteria into subtasks:", "`agent-ops task create --parent <task-id>` records one, and", "`agent-ops task status --parent <task-id>` lists them. Each subtask", "carries its own criteria and verification; final review covers all", "tasks together when their worktrees are integrated;", "completing one never completes its parent.", "", ...autoRunInstruction(descriptor, context), "To review every active subtask of a parent together, run", "`agent-ops batch --parent <task-id> --yes` instead of one review per", "subtask. It needs what review needs: start it through the trusted outer", "host runner as its first and only invocation, with the same Codex", "elevation, and keep `--output <file>` instead of a shell redirect. It", "verifies, one task at a time, each task that lacks fresh evidence, and", "reviews a few tasks at once. Do not commit or edit the worktree while it", "runs: a moved HEAD voids every verify and review in the batch.", "For subagents in separate worktrees, use `task advance` instead.", "", "Delegate a subtask whose scope spans unknown files, broad scans, or an", "independent workstream instead of doing it inline. Name the bounded", "scope, the artifact to return, and the acceptance criteria it serves.", subagentInstruction(descriptor), "Read every returned artifact back yourself: integration, verification,", "and review stay with the coordinator.", "", "One writer at a time. `agent-ops verify` and `agent-ops review`", "fingerprint the working tree and report UNKNOWN when anything else", "writes it during the run, so delegate edits to one subagent at a time,", "or give each writer its own Git worktree and merge before verifying.", "Read-only delegation has no such limit and may run in parallel.", "", "Parallel conversations in one repository each need their own worktree.", "When `.agent-ops/config.json` sets `worktree.mode` to `auto`, run", "`agent-ops task create` from the main checkout before your first edit:", "it creates this session's worktree and puts the task there (`agent-ops", "worktree add <name> --session <session-id>` does the same without a", "task). Work only inside the printed path", "(Claude Code: EnterWorktree with that path; otherwise use it as every", "command's working directory). Commit, then run verify and review for", "each task with the printed `--base`; do not run `task complete` in the", "worktree. Leave it (Claude Code: ExitWorktree with action keep) and run", "`agent-ops worktree finish <name>` from the main checkout: it completes", "every task in the worktree, subtasks first, then merges. A finish that", "reports an incomplete task names it; verify or review that task again", "in the worktree and finish again. A finish that reports a conflict is", "resolved once, as a new task whose criteria cover both intents; if its", "verification or review fails, reset as instructed and report.", "", "For a session with writing subagents, record each child task's complete", "modification intent with `task create --intent <text>` before editing.", "Each child commits and runs local `verify --task` in its own worktree.", "From the main checkout run `agent-ops task advance --task <parent-id>", "--session <session-id> --yes` through the trusted outer host runner,", "unsetting `CODEX_SANDBOX_NETWORK_DISABLED` on Codex. Advance integrates", "child commits and intent into the coordinator worktree, verifies every", "task on the final candidate, then runs one two-round tree review.", "Only `NOT_RUN / scope-too-large` falls back to per-task two-round review;", "a substantive FAIL must be fixed. After PASS, advance writes a local", "receipt under the main checkout's Git common directory and calls finish.", "If the target moves, it retries the final gate once. An unresolved", "failure preserves worktrees and cannot be reported as finished.", "");
     }
     if (context.capabilities.includes("lifecycle-summary")) {
         lines.push("Advisory lifecycle summaries and local logs are informational. Advisory", "failures must remain fail-open and cannot become verification evidence.", "");

package/dist/runtime/src/install/plan.js

@@ -79,7 +79,7 @@ async function detectVerificationCommands(root) {
         blockers: discovery.decisions.map((decision) => `${decision.adapter}: ${decision.message}`)
     };
 }
-function buildConfig(profiles, existing, reviewTargets = [], detectedCommands = [], completionGateEnabled = false, worktree) {
+function buildConfig(profiles, existing, reviewTargets = [], detectedCommands = [], completionGateEnabled = false, worktree, officeEnabled) {
     // Absent reviewRoles means external review is disabled; an empty selection
     // must therefore omit the field rather than write an empty array.
     const reviewRoles = reviewTargets.length > 0
@@ -96,13 +96,14 @@ function buildConfig(profiles, existing, reviewTargets = [], detectedCommands =
         schemaVersion: CONFIG_SCHEMA_VERSION,
         profiles: [...profiles],
         verification,
-        features: existing?.features ?? {
-            stopVerification: {
-                enabled: false
-            },
-            completionGate: {
-                enabled: completionGateEnabled
-            }
+        features: {
+            ...(existing?.features ?? {
+                stopVerification: { enabled: false },
+                completionGate: { enabled: completionGateEnabled }
+            }),
+            ...(officeEnabled === undefined
+                ? (existing === undefined ? { office: { enabled: false } } : {})
+                : { office: { enabled: officeEnabled } })
         },
         pathMappings: existing?.pathMappings ?? [],
         securityExceptions: existing?.securityExceptions ?? [],
@@ -110,7 +111,7 @@ function buildConfig(profiles, existing, reviewTargets = [], detectedCommands =
         ...(effectiveWorktree === undefined ? {} : { worktree: effectiveWorktree })
     };
 }
-async function planConfig(root, profiles, existingManifest, suppliedConfig, reviewTargets = [], completionGateEnabled = false, worktree) {
+async function planConfig(root, profiles, existingManifest, suppliedConfig, reviewTargets = [], completionGateEnabled = false, worktree, officeEnabled) {
     const current = await readCurrentFile(root, CONFIG_PATH);
     const owned = findOwnedArtifact(existingManifest, CONFIG_PATH);
     if (current !== null && owned === undefined) {
@@ -146,7 +147,7 @@ async function planConfig(root, profiles, existingManifest, suppliedConfig, revi
         existingConfig.verification.commands.length === 0
         ? await detectVerificationCommands(root)
         : { commands: [], blockers: [] };
-    const config = buildConfig(profiles, existingConfig, reviewTargets, detected.commands, completionGateEnabled, worktree);
+    const config = buildConfig(profiles, existingConfig, reviewTargets, detected.commands, completionGateEnabled, worktree, officeEnabled);
     const content = `${JSON.stringify(config, null, 2)}\n`;
     return {
         operation: {
@@ -437,6 +438,10 @@ export async function createInstallPlan(options) {
     await assertCodexLoopConfiguration(options.root, options.harness, resolved.capabilities);
     const existing = await readExistingManifest(options.root);
     assertCompatibleManifest(existing?.manifest ?? null, options.scope, options.harness, options.allowHarnessChange === true);
+    const config = await planConfig(options.root, resolved.profiles, existing?.manifest ?? null, options.existingConfig, options.reviewTargets ?? [], completionGateEnabled, options.worktree, options.officeEnabled);
+    resolved.capabilities = resolved.capabilities.filter(capability => capability !== "office-presence");
+    if (config.config.features.office?.enabled === true)
+        resolved.capabilities.push("office-presence");
     const existingOpencodePluginPath = existing?.manifest.artifacts.find(({ id }) => id === "opencode-plugin")?.path;
     const explicitHookTargets = new Map();
     for (const target of options.hookTargets ?? []) {
@@ -520,7 +525,6 @@ export async function createInstallPlan(options) {
         : [];
     const operations = [];
     const artifacts = [];
-    const config = await planConfig(options.root, resolved.profiles, existing?.manifest ?? null, options.existingConfig, options.reviewTargets ?? [], completionGateEnabled, options.worktree);
     operations.push(config.operation);
     artifacts.push(config.record);
     const repaired = [];
@@ -634,6 +638,7 @@ export async function createInstallPlan(options) {
         scope: options.scope,
         harness: options.harness,
         desired: config.config.worktree?.mode === "auto",
+        run: resolved.profiles.includes("run"),
         ...(options.codexHome === undefined ? {} : { codexHome: options.codexHome })
     });
     return {

package/dist/runtime/src/install/preauth.js

@@ -40,6 +40,9 @@ export const CLAUDE_PREAUTH_UNSANDBOXED = [
     "npx lint-staged",
     "npx lint-staged *"
 ];
+/** Added with the `run` profile: `agent-ops run` starts a launchd supervisor. */
+export const CLAUDE_PREAUTH_RUN_ALLOW = ["Bash(agent-ops run *)"];
+export const CLAUDE_PREAUTH_RUN_UNSANDBOXED = ["agent-ops run", "agent-ops run *"];
 export const CODEX_RULES_PATH = "rules/agent-ops.rules";
 export const CODEX_RULES_MARKER = "# agent-ops managed: regenerated by agent-ops init and update; do not edit.";
 const HOSTS = ["AGENT_OPS_HOST=codex", "AGENT_OPS_HOST=claude", "AGENT_OPS_HOST=agy"];
@@ -129,6 +132,23 @@ export const CODEX_RULES_CONTENT = `${[
         ];
     })
 ].join("\n")}\n`;
+/**
+ * The `run` profile's rule: one argv prefix covers bare `agent-ops run` and
+ * every `agent-ops run *` shape the rules tell Codex to start escalated.
+ */
+export const CODEX_RUN_RULE = rule({
+    withHost: false,
+    command: ["run"],
+    justification: "agent-ops run starts a launchd supervisor and reviews, which need network and loopback outside the sandbox.",
+    match: "env -u CODEX_SANDBOX_NETWORK_DISABLED agent-ops run --goal-file .agent-ops/state/run-goal.md --host codex --wait",
+    notMatch: "env -u CODEX_SANDBOX_NETWORK_DISABLED agent-ops trust grant --scope project --yes"
+});
+/** The rules file for worktree auto mode and the `run` profile, or null for neither. */
+export function codexRulesContent(worktree, run) {
+    if (!run)
+        return worktree ? CODEX_RULES_CONTENT : null;
+    return `${worktree ? CODEX_RULES_CONTENT : `${CODEX_RULES_MARKER}\n`}${CODEX_RUN_RULE}\n`;
+}
 export function codexHomeDirectory(env = process.env) {
     return env.CODEX_HOME ?? join(env.HOME ?? homedir(), ".codex");
 }
@@ -165,7 +185,7 @@ function stringList(value, where) {
  * settings. Nothing else in the file is touched. `null` content means the file
  * held nothing but these entries and should go; `undefined` means no change.
  */
-export function planClaudeLocalSettings(source, desired) {
+export function planClaudeLocalSettings(source, desired, run = false) {
     let settings = {};
     if (source !== null && source.trim() !== "") {
         let parsed;
@@ -188,11 +208,14 @@ export function planClaudeLocalSettings(source, desired) {
     }
     const allow = stringList(permissions.allow, "permissions.allow");
     const excluded = stringList(sandbox.excludedCommands, "sandbox.excludedCommands");
-    const merge = (current, managed) => desired
-        ? [...current, ...managed.filter((entry) => !current.includes(entry))]
-        : current.filter((entry) => !managed.includes(entry));
-    const nextAllow = merge(allow, CLAUDE_PREAUTH_ALLOW);
-    const nextExcluded = merge(excluded, CLAUDE_PREAUTH_UNSANDBOXED);
+    // Run entries leave with the profile, so an update without it strips them.
+    const merge = (current, managed, runManaged) => {
+        const wanted = [...(desired ? managed : []), ...(run ? runManaged : [])];
+        const kept = current.filter((entry) => wanted.includes(entry) || ![...managed, ...runManaged].includes(entry));
+        return [...kept, ...wanted.filter((entry) => !kept.includes(entry))];
+    };
+    const nextAllow = merge(allow, CLAUDE_PREAUTH_ALLOW, CLAUDE_PREAUTH_RUN_ALLOW);
+    const nextExcluded = merge(excluded, CLAUDE_PREAUTH_UNSANDBOXED, CLAUDE_PREAUTH_RUN_UNSANDBOXED);
     if (nextAllow.length === allow.length && nextExcluded.length === excluded.length &&
         nextAllow.every((entry, index) => entry === allow[index]) &&
         nextExcluded.every((entry, index) => entry === excluded[index])) {
@@ -218,15 +241,16 @@ export function planClaudeLocalSettings(source, desired) {
         content: Object.keys(next).length === 0 ? null : `${JSON.stringify(next, null, 2)}\n`
     };
 }
-async function claudeChange(root, desired) {
+async function claudeChange(root, desired, run) {
     const current = await readCurrent(root, CLAUDE_LOCAL_SETTINGS_PATH);
-    const planned = planClaudeLocalSettings(current?.content ?? null, desired);
+    const planned = planClaudeLocalSettings(current?.content ?? null, desired, run);
     if (planned === undefined) {
         return undefined;
     }
-    const summary = desired
+    const summary = desired || run
         ? `Adds to ${CLAUDE_LOCAL_SETTINGS_PATH} (existing entries kept): permissions.allow ` +
-            `${JSON.stringify(CLAUDE_PREAUTH_ALLOW)}; sandbox.excludedCommands ${JSON.stringify(CLAUDE_PREAUTH_UNSANDBOXED)}.`
+            `${JSON.stringify([...(desired ? CLAUDE_PREAUTH_ALLOW : []), ...(run ? CLAUDE_PREAUTH_RUN_ALLOW : [])])}; sandbox.excludedCommands ` +
+            `${JSON.stringify([...(desired ? CLAUDE_PREAUTH_UNSANDBOXED : []), ...(run ? CLAUDE_PREAUTH_RUN_UNSANDBOXED : [])])}.`
         : `Removes the agent-ops entries from ${CLAUDE_LOCAL_SETTINGS_PATH}, keeping everything else.`;
     if (planned.content === null) {
         return current === null
@@ -250,7 +274,8 @@ async function claudeChange(root, desired) {
         ownedContent: false
     };
 }
-async function codexChange(codexHome, desired) {
+async function codexChange(codexHome, desired, run) {
+    const content = codexRulesContent(desired, run);
     const isDirectory = await lstat(codexHome).then((status) => status.isDirectory(), () => false);
     if (!isDirectory) {
         return undefined;
@@ -261,7 +286,7 @@ async function codexChange(codexHome, desired) {
     if (current !== null && !current.content.startsWith(CODEX_RULES_MARKER)) {
         return undefined;
     }
-    if (!desired) {
+    if (content === null) {
         return current === null
             ? undefined
             : {
@@ -271,37 +296,38 @@ async function codexChange(codexHome, desired) {
                 ownedContent: true
             };
     }
-    return current?.content === CODEX_RULES_CONTENT
+    return current?.content === content
         ? undefined
         : {
             root: codexHome,
             operation: {
                 kind: "write",
                 path: CODEX_RULES_PATH,
-                content: CODEX_RULES_CONTENT,
+                content,
                 expectedHash: current?.hash ?? null
             },
-            summary: `Writes ${join(codexHome, CODEX_RULES_PATH)}: allow rules for agent-ops review, batch, doctor --check-auth, worktree commit and lint-staged.`,
+            summary: `Writes ${join(codexHome, CODEX_RULES_PATH)}: allow rules for ${desired ? `agent-ops review, batch, ${run ? "run, " : ""}doctor --check-auth, worktree commit and lint-staged` : "agent-ops run"}.`,
             ownedContent: true
         };
 }
 /**
  * What init, update and uninstall change beyond the manifest: the local Claude
  * settings entries and the Codex rules file. Project scope only; `desired`
- * false plans the removal of what agent-ops wrote.
+ * (worktree auto mode) and `run` false plan the removal of what agent-ops wrote.
  */
 export async function planPreauth(options) {
+    const run = options.run === true;
     if (options.scope !== "project") {
         return [];
     }
     const changes = [];
     if (options.harness.includes("claude")) {
-        changes.push(await claudeChange(options.root, options.desired));
+        changes.push(await claudeChange(options.root, options.desired, run));
     }
     // Codex's home is outside the project, so it is only touched when the caller
     // names it (the CLI does): planning alone must never reach into a real home.
     if (options.harness.includes("codex") && options.codexHome !== undefined) {
-        changes.push(await codexChange(options.codexHome, options.desired));
+        changes.push(await codexChange(options.codexHome, options.desired, run));
     }
     return changes.filter((change) => change !== undefined);
 }

package/dist/runtime/src/install/profiles.js

@@ -1,16 +1,20 @@
 import { AgentOpsError } from "../fs/paths.js";
-const PROFILE_ORDER = ["core", "advisory", "guardrails", "loop"];
+const PROFILE_ORDER = ["core", "advisory", "guardrails", "loop", "run"];
 export const PROFILE_CAPABILITIES = {
     core: ["rules", "task", "verify", "review"],
     advisory: ["lifecycle-summary", "local-log"],
     guardrails: ["command-policy"],
-    loop: ["project-loop"]
+    loop: ["project-loop"],
+    run: ["auto-run"]
 };
 export function resolveProfiles(inputProfiles) {
     if (inputProfiles.length === 0) {
         throw new AgentOpsError("PROFILE_REQUIRED", "At least one installation profile is required.");
     }
     const selectedProfiles = new Set(inputProfiles);
+    if (selectedProfiles.has("run")) {
+        selectedProfiles.add("loop");
+    }
     if (selectedProfiles.has("guardrails") ||
         selectedProfiles.has("loop")) {
         selectedProfiles.add("core");
@@ -30,6 +34,8 @@ export function resolveProfiles(inputProfiles) {
 }
 export function resolveCapabilities(config) {
     const resolved = resolveProfiles(config.profiles);
+    if (config.features.office?.enabled === true)
+        resolved.capabilities.push("office-presence");
     if (config.features.stopVerification.enabled) {
         resolved.capabilities.push("optional-stop-verify");
     }

package/dist/runtime/src/install/update.js

@@ -114,6 +114,7 @@ export async function createUpdatePlan(options) {
         adapters: options.adapters,
         toolkitVersion: options.toolkitVersion ?? targetVersion,
         allowHarnessChange: true,
+        ...(options.officeEnabled === undefined ? {} : { officeEnabled: options.officeEnabled }),
         ...(options.hookRuntimePath === undefined
             ? {}
             : { hookRuntimePath: options.hookRuntimePath }),

package/dist/runtime/src/office/collect.js

@@ -0,0 +1,233 @@
+import { execFile } from "node:child_process";
+import { readdir, readFile, realpath, stat } from "node:fs/promises";
+import { basename, join, resolve } from "node:path";
+import { promisify } from "node:util";
+import { readWorktreeRecord, parseWorktreeListPorcelain } from "../parallel/service.js";
+import { FileRunRepository } from "../run/service.js";
+import { criteriaProgress } from "../run/phase.js";
+import { parseTaskStateSource } from "../task/store.js";
+import { findReviewAttestation } from "../review/attestation.js";
+import { FileEvidenceStore } from "../verify/evidence.js";
+import { readPrivateFile } from "../security/permissions.js";
+import { OFFICE_SESSION_RETENTION_MS, readOfficeSessions } from "./sessions.js";
+const run = promisify(execFile);
+export const officeGit = async (cwd, args) => {
+    try {
+        return { exitCode: 0, stdout: (await run("git", [...args], { cwd, maxBuffer: 8 * 1024 * 1024 })).stdout };
+    }
+    catch {
+        return { exitCode: 1, stdout: "" };
+    }
+};
+const MAX_PATHS = 200;
+const OFFICE_PHASES = new Set(["planning", "implementing", "verifying", "reviewing", "integrating", "unknown"]);
+function officePhase(value) {
+    return value === undefined ? undefined : OFFICE_PHASES.has(value) ? value : "unknown";
+}
+function evidenceValue(value, taskId, criterionId) {
+    if (typeof value !== "object" || value === null || Array.isArray(value))
+        return null;
+    const evidence = value;
+    return evidence.taskId === taskId && evidence.criterionId === criterionId &&
+        (evidence.status === "PASS" || evidence.status === "FAIL" || evidence.status === "UNKNOWN")
+        ? value : null;
+}
+async function taskView(root, record) {
+    const store = new FileEvidenceStore(root, root);
+    const loaded = [];
+    for (const criterion of record.task.criteria) {
+        for (const reference of record.evidence[criterion.id] ?? []) {
+            if (reference.startsWith("review:"))
+                continue;
+            const item = await store.load(reference).catch(() => null);
+            const parsed = evidenceValue(item, record.task.id, criterion.id);
+            if (parsed !== null)
+                loaded.push({ reference, evidence: parsed });
+        }
+    }
+    // A task keeps its evidence history. The desk shows the newest observation
+    // for each verifier/acceptance phase so an old failure cannot poison a
+    // later PASS, and a stale review cannot remain current after a new verify.
+    const latest = new Map();
+    for (const item of loaded) {
+        const phase = item.evidence.acceptance?.phase ?? "command";
+        const key = `${item.evidence.criterionId}\0${item.evidence.commandId}\0${phase}`;
+        const previous = latest.get(key);
+        const currentAt = Date.parse(item.evidence.finishedAt);
+        const previousAt = previous === undefined ? Number.NEGATIVE_INFINITY : Date.parse(previous.evidence.finishedAt);
+        if (previous === undefined || currentAt >= previousAt)
+            latest.set(key, item);
+    }
+    const evidence = [...latest.values()].map(item => item.evidence);
+    let passed = 0;
+    for (const criterion of record.task.criteria) {
+        const rows = evidence.filter(item => item.criterionId === criterion.id);
+        const result = criteriaProgress([criterion], {
+            results: rows.map(item => ({ status: item.status })),
+            acceptance: rows.flatMap(item => item.acceptance?.checks.map(check => ({ criterionId: criterion.id, status: check.status })) ?? [])
+        });
+        passed += result.passed;
+    }
+    const total = record.task.criteria.length;
+    const verify = evidence.some(item => item.status === "FAIL" || item.status === "UNKNOWN")
+        ? "FAIL"
+        : total > 0 && passed === total
+            ? "PASS"
+            : null;
+    const newest = evidence.reduce((current, item) => current === null || Date.parse(item.finishedAt) >= Date.parse(current.finishedAt) ? item : current, null);
+    const review = newest !== null &&
+        await findReviewAttestation(root, newest.sourceFingerprint, record.task.id) !== null ? "PASS" : null;
+    const progress = { verify, review, passed, total };
+    const phase = record.status === "complete"
+        ? "integrating"
+        : review === "PASS"
+            ? "reviewing"
+            : verify !== null
+                ? "verifying"
+                : evidence.length > 0
+                    ? "implementing"
+                    : "planning";
+    return { taskId: record.task.id, status: record.status, phase, progress, completedAt: record.completedAt };
+}
+async function taskState(root) {
+    const source = await readPrivateFile(join(root, ".agent-ops", "tasks", "state.json"), root).catch(() => null);
+    if (source === null)
+        return null;
+    try {
+        return parseTaskStateSource(source);
+    }
+    catch {
+        return null;
+    }
+}
+async function attachedTaskView(root, sessionId, fallbackTaskId) {
+    const state = await taskState(root);
+    if (state === null)
+        return null;
+    const taskId = state.sessions.find(session => session.sessionId === sessionId)?.taskId ?? fallbackTaskId;
+    if (taskId === undefined)
+        return null;
+    const record = state.tasks.find(task => task.task.id === taskId);
+    return record === undefined ? null : await taskView(root, record);
+}
+/** Changed paths and line counts against the worktree's base, committed or not. */
+export async function worktreeDiff(git, path, base) {
+    // The record is a file inside the checkout; only a commit id may reach git's argument list.
+    const numstat = /^[0-9a-f]{40,64}$/u.test(base) ? await git(path, ["diff", "--numstat", "-z", base, "--"]) : { exitCode: 1, stdout: "" };
+    const untracked = await git(path, ["ls-files", "--others", "--exclude-standard", "-z"]);
+    let insertions = 0, deletions = 0;
+    const paths = [];
+    // -z numstat: "added\tdeleted\tpath\0", renames as "added\tdeleted\t\0from\0to\0".
+    const fields = numstat.stdout.split("\0");
+    for (let i = 0; i < fields.length; i++) {
+        const match = /^(\d+|-)\t(\d+|-)\t(.*)$/u.exec(fields[i]);
+        if (match === null)
+            continue;
+        insertions += Number(match[1]) || 0;
+        deletions += Number(match[2]) || 0;
+        paths.push(match[3] === "" ? (i += 2, fields[i]) : match[3]);
+    }
+    paths.push(...untracked.stdout.split("\0").filter(p => p !== ""));
+    let recent = null, newest = -1;
+    for (const p of paths.slice(0, MAX_PATHS)) {
+        const mtime = await stat(join(path, p)).then(s => s.mtimeMs, () => -1);
+        if (mtime > newest) {
+            newest = mtime;
+            recent = p;
+        }
+    }
+    return { files: paths.length, insertions, deletions, paths: paths.slice(0, MAX_PATHS), recent: recent ?? paths[0] ?? null };
+}
+/** Each run is read on its own: one unreadable run must not hide the building. */
+export async function readRuns(commonDir) {
+    const directory = join(commonDir, "agent-ops", "runs");
+    const repository = new FileRunRepository(directory, commonDir);
+    const names = await readdir(directory).catch(() => []);
+    const runs = [];
+    for (const name of names.filter(n => n.endsWith(".json"))) {
+        const state = await repository.read(basename(name, ".json")).catch(() => null);
+        if (state !== null)
+            runs.push(state);
+    }
+    return runs.sort((a, b) => a.createdAt.localeCompare(b.createdAt));
+}
+async function readSessions(mainRoot, commonDir, now, trustedRoots = [mainRoot]) {
+    const sessions = await readOfficeSessions(commonDir, now).catch(() => []);
+    const views = await Promise.all(sessions.map(async (session) => {
+        let task = null;
+        for (const root of trustedRoots) {
+            task = await attachedTaskView(root, session.sessionId, root === mainRoot ? session.taskId : undefined);
+            if (task !== null)
+                break;
+        }
+        const observedPhase = session?.phase === undefined ? undefined : officePhase(session.phase) ?? "unknown";
+        const phase = task === null
+            ? observedPhase
+            : task.status === "complete"
+                ? task.phase
+                : observedPhase ?? task.phase;
+        return task === null
+            ? session
+            : { ...session, taskId: task.taskId, phase, taskStatus: task.status, progress: task.progress,
+                completedAt: task.completedAt, status: task.status === "complete" ? "idle" : session.status };
+    }));
+    return views.filter(session => session.completedAt === undefined || session.completedAt === null ||
+        now - Date.parse(session.completedAt) <= OFFICE_SESSION_RETENTION_MS);
+}
+function alive(pid) {
+    try {
+        process.kill(pid, 0);
+        return true;
+    }
+    catch (error) {
+        return error.code === "EPERM";
+    }
+}
+/** Review slots held by a live process (see review/slots.ts). */
+export async function readReviewSlots(commonDir) {
+    const slots = [];
+    for (const name of (await readdir(commonDir).catch(() => [])).sort()) {
+        const match = /^agent-ops-review-slot-(\d+)\.lock$/u.exec(name);
+        if (match === null)
+            continue;
+        try {
+            const owner = JSON.parse(await readFile(join(commonDir, name, "owner.json"), "utf8"));
+            if (typeof owner.pid !== "number" || typeof owner.at !== "number" || !alive(owner.pid))
+                continue;
+            slots.push({ slot: Number(match[1]), since: new Date(owner.at).toISOString(),
+                taskId: typeof owner.taskId === "string" ? owner.taskId : null, root: typeof owner.root === "string" ? owner.root : null });
+        }
+        catch { /* a slot being claimed or released */ }
+    }
+    return slots;
+}
+export async function collectOfficeInput(mainRoot, commonDir, git = officeGit, now = Date.now()) {
+    const worktrees = [];
+    const discovered = [];
+    for (const entry of parseWorktreeListPorcelain((await git(mainRoot, ["worktree", "list", "--porcelain"])).stdout)) {
+        const path = await realpath(resolve(entry.path)).catch(() => resolve(entry.path));
+        const record = await readWorktreeRecord(path).catch(() => null);
+        if (record === null)
+            continue;
+        discovered.push({ path, record });
+    }
+    const sessions = await readSessions(mainRoot, commonDir, now, [mainRoot, ...discovered.map(item => item.path)]);
+    for (const { path, record } of discovered) {
+        const session = sessions.find(item => item.sessionId === record.sessionId);
+        const task = await attachedTaskView(path, record.sessionId, session?.taskId);
+        const base = { name: record.name, path, branch: record.branch, sessionId: record.sessionId,
+            ...(record.runId === undefined ? {} : { runId: record.runId }), diff: await worktreeDiff(git, path, record.base) };
+        const observedPhase = session?.phase === undefined ? undefined : officePhase(session.phase) ?? "unknown";
+        const phase = task === null
+            ? observedPhase
+            : task.status === "complete"
+                ? task.phase
+                : observedPhase ?? task.phase;
+        worktrees.push(task === null
+            ? session === undefined ? base : { ...base, ...(officePhase(session.phase) === undefined ? {} : { phase: officePhase(session.phase) }), status: session.status,
+                ...(session.taskId === undefined ? {} : { taskId: session.taskId }), host: session.host ?? session.harness }
+            : { ...base, phase, status: task.status, taskId: task.taskId, progress: task.progress,
+                ...(task.completedAt === null ? {} : { completedAt: task.completedAt }), host: session?.host ?? session?.harness });
+    }
+    return { runs: await readRuns(commonDir), worktrees, sessions, reviews: await readReviewSlots(commonDir), now };
+}