@phnx-labs/agents-cli 1.20.72 → 1.20.74

package/CHANGELOG.md

@@ -1,5 +1,400 @@
 # Changelog
 
+## 1.20.74
+
+- **Project resource manifests are now portable across Windows and POSIX.** The
+  managed-resource manifest `.agents-managed.json` recorded its paths with the
+  host's native separator, so a sync run on Windows wrote entries like
+  `skills\myskill`. That file lives in the version-controlled project `.agents`
+  dir and travels between machines, and the cleanup pass matches manifest entries
+  with `path.sep` — so a manifest written on Windows silently failed to match on
+  macOS or Linux and left previously managed files behind on the next sync (and
+  vice versa). Manifest paths are now normalized to POSIX separators on write and
+  on read, which also repairs manifests written by earlier Windows builds. Source:
+  `apps/cli/src/lib/project-resources.ts`.
+
+- **`agents import <agent> --isolated` — bring your existing setup into a sandbox.**
+  Isolation was a cold start: a new isolated copy began empty, and the only way to get
+  settings into it was by hand. A plain `agents import` is the opposite of what is
+  wanted here — it *adopts*, moving `~/.<agent>` into a version home, symlinking the
+  original away, setting the global default and creating a shim (and is now refused
+  outright for an isolated-only agent). `--isolated` copies instead: your settings land
+  in the isolated home, your real config stays exactly where it is, and the version is
+  finalized the way `agents add --isolated` does — versioned alias and marker, no
+  default, no bare shim, no config symlink. Credentials are skipped by default and
+  named in the output rather than silently included, since an isolated copy signs in as
+  its own principal; `--with-auth` opts in. Symlinks into `~/.agents` are dropped so the
+  copy does not depend on the CLI's tree. Source: `apps/cli/src/lib/import.ts`,
+  `apps/cli/src/commands/import.ts`.
+
+- **`agents use <agent>@<isolated>` now works, and a bare `agents run <agent>` reaches
+  your isolated copy.** Isolated installs were unreachable by name: `resolveVersion`
+  ended at the global default, and an isolated install deliberately never becomes one —
+  so `agents use` refused, and an isolated-only user had to type the full
+  `agents run codex@0.144.6` every time while a bare `agents run codex` fell through to
+  whatever `codex` meant on PATH. `use` now records an **isolated default** instead of
+  refusing, and resolution falls back to it (`project pin -> global default -> isolated
+  default`). Strictly a fallback, so nothing changes for anyone who has a global
+  default. The pointer lives in `isolatedAgents:` in `agents.yaml`, never in the global
+  `agents:` map — that separation is what keeps `getGlobalDefault` incapable of
+  returning an isolated version, and with it the launcher, bare shim, config symlink and
+  self-heal `shadowing` check all stay out of reach. It is verified on read and
+  re-pointed (or cleared) on removal, so it can never resolve to a version that is gone.
+  `agents view` labels it `(isolated default)`. Source: `apps/cli/src/lib/versions.ts`,
+  `apps/cli/src/commands/versions.ts`, `apps/cli/src/commands/view.ts`.
+
+- **`agents export <agent>[@<version>]` — take an isolated install's config with you.**
+  `--isolated` was a one-way door: it builds a self-contained home under the version
+  dir and nothing ever brings that work back, so a user who configured a sandboxed copy
+  for a week had to copy files by hand to promote it — or to leave. Export is additive
+  by default: it copies only paths you don't already have, and a collision is **not**
+  silently skipped — the incoming file is written beside yours as
+  `<name>.from-agents-cli` so you can `--diff` it and take the parts you want. Your
+  files are never modified. `--replace` promotes a sandbox wholesale (yours is moved to
+  `backups/<agent>/<ts>`, and it is the only mode that asks for confirmation);
+  `--staged` dumps the tree into `~/.<agent>/.agents-export-<ts>/` and activates
+  nothing. Every mode strips symlinks pointing back into `~/.agents` so the result
+  keeps working after agents-cli is gone, keeps your own symlinks, and writes a receipt
+  to `~/.<agent>/.agents-cli-export.json` recording exactly what came from the export —
+  which makes "which of these files are mine?" answerable and the whole thing
+  reversible. A `~/.<agent>` that agents-cli already adopted is refused, since writing
+  there would mutate that version's home rather than your config. File *contents* are
+  never auto-merged: the TOML parser here drops comments across parse+stringify, so
+  unioning keys would silently delete them. Source: `apps/cli/src/lib/export.ts`,
+  `apps/cli/src/commands/export.ts`, `apps/cli/src/lib/config-transfer.ts`.
+
+- **An isolated-only agent can no longer be adopted by anything.** `--isolated` used to
+  be defined by what it *doesn't* do — no global default, no bare shim, no config
+  symlink, no PATH edit — which meant every code path that could adopt an agent had to
+  remember to check first. It leaked three times that way. Protection is now derived
+  from the `.isolated` markers on disk (`isIsolationProtected`: at least one installed
+  version, and every one isolated) and enforced inside the five primitives that can
+  cross the boundary — `setGlobalDefault`, `createShim`, `switchConfigSymlink`,
+  `switchHomeFileSymlinks`, `adoptShadowingLauncher` — so refusal is a property of the
+  code rather than a convention. There is no mode to set and none to forget: installing
+  with `--isolated` *is* the opt-in, it is per-agent, and the escape hatch is inherent
+  (remove the isolated copies and the agent is ordinary again). `agents add`,
+  `agents import` and `doctor --adopt` refuse with guidance rather than a stack trace —
+  `import` is additionally checked at its entry point, because it registers the adopted
+  install as a normal version *before* adopting, which would otherwise un-protect the
+  agent underneath the primitive gate. Clearing a global default stays allowed, since
+  removal legitimately clears one as an agent becomes isolated-only. A completeness test
+  pins the primitive list and scans for any new ungated mutator. Source:
+  `apps/cli/src/lib/shims.ts`, `apps/cli/src/lib/versions.ts`,
+  `apps/cli/src/lib/isolation-boundary-report.ts`.
+
+- **`agents view` no longer hides your own CLI behind an isolated install.** The listing
+  was either/or per agent: any managed version at all suppressed the "Not Managed by
+  Agents CLI" block, so a single `agents add <agent>@<v> --isolated` made the user's
+  globally-installed CLI disappear from the one command they'd run to confirm
+  `--isolated` had left it alone. Nothing on disk was ever touched — the isolation
+  boundary holds — but the report read exactly like the damage it was supposed to rule
+  out. Isolated copies now render alongside the global install and are tagged
+  `9.9.4 (isolated)`; a normal (non-isolated) version still takes the launcher over and
+  still suppresses the global row, since that row would just be our own shim. The global
+  row is also resolved from PATH now (`getUnmanagedCliState`) instead of from the version
+  dirs, which could otherwise report an isolated copy — deliberately unreachable from
+  PATH — as `(global)`. Source: `apps/cli/src/commands/view.ts`, `apps/cli/src/lib/agents.ts`.
+
+## 1.20.73
+
+- **`agents cli install <binary-cli>` no longer hardcodes `/usr/local/bin` (#1103).**
+  Binary-method installs downloaded (and extracted archives) straight into
+  `/usr/local/bin`, which fails with `EACCES` on Apple Silicon Macs where that
+  directory is root-owned and not user-writable. A new `resolveBinDir()` picks
+  the install directory instead: honor `AGENTS_CLI_BIN_DIR` if set, else prefer
+  `~/.local/bin` (created on demand — the same XDG user-bin dir shims already
+  use), else fall back to `/usr/local/bin` with an actionable error pointing at
+  `AGENTS_CLI_BIN_DIR` / `~/.local/bin` instead of a bare `EACCES`. Source:
+  `apps/cli/src/lib/cli-resources.ts`, `apps/cli/src/lib/cli-resources.test.ts`.
+
+- **Stream host follows over one persistent SSH connection (RUSH-1407).** `run --host` and `hosts logs -f` now follow remote logs with a long-lived `tail -f` stream that reconnects from the saved byte offset and captures the remote `.exit` code without per-cycle SSH spawns. Source: `apps/cli/src/lib/hosts/progress.ts`.
+
+- **Reuse warm crabbox boxes from `agents run` (RUSH-1609).** `agents run <agent> "<task>" --box <slug>` now targets an existing warm crabbox box, runs the same bootstrap and credential provisioning as `--lease`, and leaves the box running for reuse across repositories. Source: `apps/cli/src/commands/exec.ts`, `apps/cli/src/lib/crabbox/lease.ts`.
+
+- **Show reasoning in Factory progress timelines (RUSH-1634).** Factory detail panes now interleave assistant prose and reasoning summaries with tool calls in the Progress rail, so agent activity explains intent instead of showing only file/tool touches. Source: `apps/factory/src/core/session.summary.ts`, `apps/factory/ui/settings/components/mission-control/Timeline.tsx`.
+
+- **Expose an Agents HQ floor snapshot bridge (RUSH-1638).** `agents hq floor --json` now emits a machine-readable floor snapshot that joins live sessions, teams, feed blocks, room placement, ambient events, and command-backed actions for HQ clients. Source: `apps/cli/src/commands/hq.ts`, `apps/cli/src/lib/hq/floor.ts`.
+
+- **Menu-bar Quick Dispatch attaches selected screenshots after ticket creation (RUSH-1693).** The helper now uploads every selected quick-capture screenshot to the created Linear issue itself after parsing the `Created RUSH-###` result, instead of relying on the ticket agent to run a second proof-upload command from its prompt. Source: `apps/cli/menubar/Sources/MenubarHelper/AgentsCLI.swift`, `apps/cli/menubar/Sources/MenubarHelper/IssueSelfTest.swift`, `apps/cli/docs/menubar.md`.
+
+- **Move supported OpenClaw secrets into Keychain-backed refs (RUSH-175).** `agents secrets openclaw-keychain migrate` now stores supported OpenClaw plaintext credentials in macOS Keychain, rewrites OpenClaw config fields to exec SecretRefs, and refuses to delete top-level env secrets that have no supported SecretRef target. Source: `apps/cli/src/lib/openclaw-keychain.ts`, `apps/cli/src/commands/secrets.ts`, `apps/cli/docs/secrets.md`.
+
+- **Provision share workers completely (RUSH-1792).** `agents share setup` now configures the R2 lifecycle rule and sets the Worker `WRITE_TOKEN` through Cloudflare's Workers Secrets API after deploying the R2-bound Worker. Source: `apps/cli/src/lib/share/provision.ts`.
+
+- **Extract a reusable share-publish endpoint seam (RUSH-1794).** `agents share <file>` now delegates its authenticated PUT (bearer token, `--slug`, `--expire`) through `publishToEndpoint`, decoupled from config/keychain loading, with a real-HTTP test asserting the wire contract. Source: `apps/cli/src/lib/share/publish.ts`.
+
+- **Map the default share domain automatically (RUSH-1796).** `agents share setup` now maps `share.agents-cli.sh` when the Cloudflare token can see the `agents-cli.sh` zone, while keeping the workers.dev endpoint when it cannot and honoring `--domain` overrides. Source: `apps/cli/src/commands/share.ts`, `apps/cli/docs/share.md`.
+
+- **Expire shared artifacts end to end (RUSH-1797).** `agents share --expire` now stores
+  `expires-at` metadata for the Worker to enforce and `agents share setup` installs a
+  managed R2 lifecycle rule so old share objects self-clean. Source:
+  `apps/cli/src/lib/share/{publish,provision,worker-template}.ts`.
+
+- **`agents share` central mode now follows synced config plus injected write tokens (RUSH-1798).** `agents share join` can bind an existing synced endpoint without reprovisioning, publish reads `SHARE_WRITE_TOKEN` from runtime env before falling back to the local `share` bundle, and agent/team/supported cloud launches propagate the token when it is already available so ephemeral agents can publish durable links with no Cloudflare setup. Source: `apps/cli/src/commands/share.ts`, `apps/cli/src/lib/share/config.ts`, `apps/cli/src/commands/exec.ts`, `apps/cli/src/commands/cloud.ts`, `apps/cli/src/commands/teams.ts`, `apps/cli/src/lib/cloud/rush.ts`, `apps/cli/src/lib/cloud/factory.ts`, `apps/cli/src/lib/cloud/codex.ts`.
+
+- **Plan-render auto-publish plumbing (RUSH-1799).** `agents share <file> --json`
+  now emits a stable `{ url, coverUrl, expiresAt }` result so plan-render hooks can
+  publish rendered HTML and post the returned link without scraping terminal output.
+  Source: `apps/cli/src/commands/share.ts`, `apps/cli/src/lib/share/publish.ts`.
+
+- Added `agents share` regression coverage for token storage, publish upload headers, expiry metadata, and Cloudflare provisioning request shapes without calling real Cloudflare in CI. Source: `src/lib/share/{config,publish-file,provision}.test.ts`, `src/lib/share/provision.ts`.
+
+- `agents repo pull` now fast-forwards the local checkout after fetch
+  (`--ff-only` semantics) and reports when it is blocked by local changes or
+  local commits instead of leaving the checkout behind origin. The system repo
+  uses the same fast-forward path as user and extra repos.
+
+- **`agents share` default links are much harder to guess (RUSH-1821).** The random
+  tail of an auto-generated share slug is now a 64-bit nonce (`randomBytes(8)`, 16 hex
+  chars) instead of the old 24-bit / 6-hex tail — closing a `~16.7M`-possibility space
+  that was small enough to brute-force. Since share reads are public (the URL is the only
+  capability), the nonce is the whole defense, so it now carries the full 64 bits.
+  Passed-in `--slug` values and existing links are unchanged. `docs/share.md` now states
+  the security model explicitly (unlisted-not-secret; reads are public; use `--expire`
+  for sensitive content; an opt-in auth-gated read is a future option). Source:
+  `apps/cli/src/lib/share/publish.ts` (`defaultSlug`), `apps/cli/src/lib/share/publish.test.ts`,
+  `apps/cli/docs/share.md`.
+
+- **`agents secrets list`/`view` gain `--json` (RUSH-1834).** Agents can now discover which secrets bundles and keys exist as machine-readable JSON before injecting one — `list --json` emits bundle metadata (name, key count, policy, backend, timestamps) and `view <bundle> --json` lists each key with its kind and stored/missing state. Values stay `null` unless `--reveal` (which keeps the same non-TTY `--plaintext` gate and audit event as the human view), so the discovery surface never leaks a secret. Gated on the explicit `--json` flag, not `stdout.isTTY`. Source: `apps/cli/src/commands/secrets.ts`.
+
+- **Per-user URL namespaces + privacy-first analytics for `agents share` (RUSH-1835).**
+  Shares now publish under the publisher's GitHub username (`share.agents-cli.sh/<user>/<slug>`),
+  with `/<user>` rendering a public gallery and legacy flat slugs still resolving. Every HTML
+  publish also injects a cookieless Cloudflare Web Analytics beacon (opt out with
+  `--no-analytics`). Configure the token during `agents share setup --analytics-token`, and
+  check status with `agents share status` / `agents share analytics`. Source:
+  `apps/cli/src/commands/share.ts`, `apps/cli/src/lib/share/{publish,analytics,worker-template}.ts`,
+  `apps/cli/src/lib/git.ts`, `apps/cli/docs/share.md`.
+
+- **Codex no longer breaks on macOS when its versioned `CODEX_HOME` overflows the
+  Unix-socket `SUN_LEN` limit.** Codex binds an app-server control socket at
+  `$CODEX_HOME/app-server-control/app-server-control.sock`, and macOS caps Unix
+  socket paths at 104 bytes (`SUN_LEN`). agents-cli points `CODEX_HOME` at the deep
+  versioned home (`~/.agents/.history/versions/codex/<version>/home/.codex`), which
+  for a typical user is long enough that the derived socket path exceeds 104 bytes —
+  so `codex app-server daemon start` failed with `path must be shorter than
+  SUN_LEN` and every codex spawn on macOS died (this took down every OpenClaw agent
+  on a mac-mini). Codex exposes no socket-path override and resolves symlinks before
+  binding, so a short symlink to the deep home does not help. The codex shims and
+  `buildExecEnv` now detect the overflow on macOS and relocate the home once to a
+  short real directory under `~/.agents/.codex-homes/<version>/.codex` (leaving a
+  symlink behind so the versioned path still resolves), keeping config, auth, and
+  state intact. A caller-set `CODEX_HOME` is always respected. Source:
+  `apps/cli/src/lib/codex-home.ts` (new), `apps/cli/src/lib/shims.ts`,
+  `apps/cli/src/lib/exec.ts`.
+
+- **Gate GitHub webhook routines on pull request labels (RUSH-203).** `agents routines add --on github:pull_request` and `agents cloud run --on pr` now preserve GitHub `--action` and `--label` filters, so a UX test routine can fire only when a PR receives `ux-approved`. Source: `apps/cli/src/lib/routines.ts`, `apps/cli/src/lib/triggers/webhook.ts`, `apps/cli/src/commands/routines.ts`, `apps/cli/src/commands/cloud.ts`.
+
+- **Remove the deprecated `agents daemon` command tree (RUSH-403).** The legacy `agents daemon start|stop|status|logs` aliases are gone for v2.0; use `agents routines start|stop|status|scheduler-logs` for scheduler controls. Source: `apps/cli/src/lib/startup/command-registry.ts`, `apps/cli/src/index.ts`, `apps/cli/src/lib/daemon.ts`, `apps/cli/docs/03-routines.md`.
+
+Added top-level resource profiles via `agents profile use`, filtering synced resources and secrets bundles by the active profile.
+Fixed source-qualified resource profile selectors for permission groups and workflows so `project:`, `user:`, and `system:` patterns match the real resource layer.
+Fixed `resolveResource` to fall through to lower-precedence layers when a higher-layer match is excluded by the active profile, matching `listResources` behavior.
+
+- **Support multiple accounts per secrets bundle (RUSH-668).** Secrets bundle keys now accept `BASE.account` names such as `GITHUB_USERNAME.personal`; selected account variants inject as the base env key and conflicting variants fail loud unless narrowed with `--keys`. Source: `apps/cli/src/lib/secrets/bundles.ts`, `apps/cli/docs/secrets.md`.
+
+- **Harden synced vault writes (RUSH-682).** Synced secret mutations now lock the vault across the full read-modify-write cycle and persist through an atomic rename, preventing concurrent CLI processes from losing each other's bundle updates. Source: `apps/cli/src/lib/secrets/vault.ts`.
+
+## Features
+
+- Added `agents login`, `agents logout`, and `agents whoami` plus `agents secrets create --synced` for age-encrypted synced secrets stored in `~/.agents/vault.age`.
+- Protected synced secrets vaults from accidental replacement: `agents login --create` and `agents login --join <path>` now require `--force` before replacing an existing `vault.age`, and synced bundle writes batch metadata plus stored keys into one vault update.
+- Synced vault encryption now runs without re-executing the `agents` binary, so standalone macOS installs can encrypt and decrypt vault data reliably; new vault writes also use the age library's default scrypt work factor.
+
+- **Keep project resources in workspace config (RUSH-705).** Project-scoped commands,
+  skills, subagents, and workflows now sync into the current workspace's `.<agent>/`
+  directory instead of lingering in global agent version homes. Source:
+  `apps/cli/src/lib/project-resources.ts`.
+- **Track Goose workflow subrecipes from first sync (RUSH-705).** Goose workflow
+  `.subrecipes/` directories are tracked in the ownership manifest from the first
+  sync, preventing workflows from being incorrectly skipped on later syncs. Source:
+  `apps/cli/src/lib/project-resources.ts`.
+
+- **`agents activity` — an event-sourced view of what agents *did*.** A new append-only, per-session event log (`~/.agents/.history/activity/<sessionId>.jsonl`) records agent-semantic milestones at hook time (plans, PRs, worktrees, sub-agents, artifacts) with no transcript re-parsing. `agents activity` renders the stream newest-first (milestones individually, routine edits collapsed to a count); `agents feed` gains a compact recent-activity lane. Source: `apps/cli/src/lib/activity.ts`, `apps/cli/src/commands/activity.ts`.
+
+- **Lease command surface: reuse picker, devices section, step UI, and Tailscale net-mode (RUSH-1922/1923/1924).**
+  Wires the command layer onto the merged crabbox-core lib:
+  - **Reuse (F3).** On an interactive `agents run … --lease`, a picker lists your
+    warm boxes (`ready` + unexpired, most-recently-touched first) and offers
+    "Provision a fresh box" / "Always provision fresh (remember for this repo)".
+    New flags: `--reuse` (scriptable — auto-pick the freshest warm box, else
+    fresh) and `--bare` (skip copying your local `~/.agents` setup onto the box,
+    i.e. `copySetup=false`). Headless / `--json` never blocks — it provisions
+    fresh unless `--reuse`/`--box` is given. New subcommands `agents lease list`
+    (`--json`) and `agents lease stop <slug>`.
+  - **Step UI (F2).** The box-side setup now renders as a live checklist —
+    each `___PHASE_<name>___` step from the lib's `onStep` stream prints via
+    `renderStepLine` (✔ Step — detail (elapsed)). Non-TTY prints one line per
+    step; `--json` emits `{phase:"setup",name,elapsedMs}` events. Host-side
+    warmup/ready/teardown phases are unchanged.
+  - **Devices (F4).** `agents devices` gains a live "Leased boxes (ephemeral ·
+    via crabbox)" section computed from `crabboxList()` — never written into the
+    device registry. `agents ssh <slug>` now resolves a leased-box slug and
+    connects to `crabbox@<tailnet-or-ip>:2222`.
+  - **Net-mode (F5).** New `--tailscale` / `--no-tailscale` on `agents run`.
+    `netMode = (--tailscale || reuse-context) && !--no-tailscale` (a solo
+    one-shot `--lease` stays public) is threaded into the lease so the lib leases
+    onto the tailnet. `agents lease setup` now also captures a Tailscale auth key
+    (EPHEMERAL, pre-authorized, `tag:crabbox`) into the `tailscale.com` secrets
+    bundle as `CRABBOX_TAILSCALE_AUTH_KEY`; when Tailscale is requested with no
+    key configured the run falls back to a public lease with an actionable hint.
+    The final "box ready/kept" line surfaces the box's tailnet FQDN/IP.
+
+  Source: `apps/cli/src/commands/exec.ts`, `apps/cli/src/commands/lease.ts`,
+  `apps/cli/src/commands/ssh.ts` (+ `*.test.ts`).
+
+- Add GitHub Copilot CLI permission sync, writing supported allow rules to `.copilot/permissions-config.json`.
+
+- **Lease lifecycle: setup-copy, step progress, and tailscale plumbing (RUSH-1920/1921/1924).** `agents run --lease` gains a library core the command layer wires up: `copySetupToBox` pushes the git-tracked subset of the local `~/.agents` onto the box and refreshes it (never `~/.claude`); the box bootstrap now echoes `___PHASE_<name>___` sentinels parsed into a structured `LeaseStep` stream (`onStep` + `renderStepLine`); and a `netMode: 'tailscale'` path leases boxes onto the tailnet (`--network tailscale -tailscale-tags tag:crabbox`, `CRABBOX_TAILSCALE_AUTH_KEY` from a secrets bundle) with `CrabboxBox.tailscaleIPv4`/`tailscaleFQDN` parsed from box labels. Source: `apps/cli/src/lib/crabbox/setup-copy.ts`, `apps/cli/src/lib/crabbox/progress.ts`, `apps/cli/src/lib/crabbox/lease.ts`, `apps/cli/src/lib/crabbox/cli.ts`.
+
+- **`agents cloud run --json` now emits machine-readable failures.** `die()` — the
+  shared fatal-exit path — always wrote red text to stderr and left stdout empty,
+  so an agent parsing `--json` output saw nothing plus a bare nonzero exit with no
+  reason. `die()` gains an optional `{ json, hint }` and, in json mode, prints
+  `{"error", "hint"?}` to stdout; a pure `formatDie()` makes the human/agent split
+  unit-testable. Every failure path in `cloud run` now threads the resolved
+  `--json` flag. Source: `apps/cli/src/lib/format.ts`, `apps/cli/src/commands/cloud.ts`.
+  (RUSH-1830)
+
+- **Show live Droid quota bars in `agents view` (RUSH-1357).** Factory billing
+  limits now render as `S`/`W`/`M` windows for Droid, matching Claude's live-usage
+  display. Source: `apps/cli/src/lib/usage.ts`.
+
+- **`agents events` is now one unified stream — operational + agent activity.** Agent-semantic events (plans, PRs, worktrees, sub-agents, artifacts) share the event vocabulary and read through the same reader as operational events (secrets, teams, commands), newest-first. `--module activity` shows agent events, `--audit` restricts to operational only, and all existing filters (`--event`, `--agent`, `--since`, `--command`) apply across both. New `readUnifiedEvents` (`apps/cli/src/lib/event-stream.ts`) is the single read surface for higher-level features. Source: `apps/cli/src/lib/events.ts`, `apps/cli/src/lib/activity.ts`, `apps/cli/src/commands/events.ts`.
+
+- **The daemon no longer crash-loops on headless Linux when a routine is overdue.**
+  On an overdue routine the daemon fires a best-effort desktop notification via
+  `notify-send` (Linux) / `osascript` (macOS). A missing notifier binary — the
+  default on a headless box without `libnotify-bin` — surfaces as an asynchronous
+  `spawn` `'error'` event, not the synchronous throw the surrounding `try/catch`
+  expected, so Node re-threw it as an uncaught exception and killed the daemon.
+  systemd then restart-looped it every ~10s, which also tore down the browser IPC
+  socket (`agents browser start` failed with "Timeout waiting for browser daemon
+  socket"). Both notifier spawns now carry an `'error'` listener so the failure is
+  swallowed as the "best-effort" contract already promised. Source:
+  `apps/cli/src/lib/overdue.ts`, `apps/cli/src/lib/overdue.test.ts`.
+
+- **`agents fleet apply` — reconcile the fleet from under the `fleet` verb.** The idempotent reconcile engine already shipped as top-level `agents apply`, but users who reach for `fleet`/`devices` as the noun (`fleet capture`, `fleet login`, `fleet status`) had no matching `fleet apply`. This surfaces the identical command as `agents fleet apply` (and `agents devices apply`) via a shared configurator, so the two can never drift — same flags, same engine, same `--plan`/`--device`/`--only` semantics. Pure discoverability alias; no behavior change to `agents apply`. Source: `apps/cli/src/commands/apply.ts` (`configureApplyCommand`, `registerFleetApplyAlias`), `apps/cli/src/commands/ssh.ts`.
+
+- **`agents fleet login` now finds the agent CLIs on the remote box.** The remote drive ran the login command over a non-login SSH shell (`ssh <box> kimi`), where the agents-cli shims (`~/.agents/.cache/shims`) are not on PATH — so `kimi`/`droid`/`codex` were "command not found" and the device-code scrape always timed out. The remote command now prepends the shim dir (resolved on the box via `$HOME`) to PATH so the login program launches. Source: `apps/cli/src/lib/fleet/remote-login.ts`.
+
+- **New `agents fleet login` — log agent CLIs into every fleet box over SSH from one browser page.** File-copying one OAuth credential across N machines is fatal: a shared refresh token rotates server-side on first refresh and invalidates the other copies. The durable fix is a per-machine login (one interactive OAuth per agent x box), and this command makes that bearable. It drives each box's device-code flow through the PTY sidecar (`ssh -tt <box> <loginCmd>`), scrapes the verification URL + user code, and surfaces every pending login in ONE local dark/light dashboard with per-code `Authorize` deep-links and TTL countdowns — so you enter codes back-to-back instead of babysitting N terminals. Default mode requests all codes concurrently; `--interactive` walks one box at a time, requesting each code just-in-time so the ~15-min TTL can't expire while you work. Only true device-code flows are driven (droid, codex, kimi); loopback / keychain-bound / uncharacterized agents (claude, gemini, antigravity, opencode, grok) are flagged non-remotable with an honest reason instead of a mis-drive. Flags: `--agents <csv>`, `--devices <csv>`, `--all`, `--interactive`, `--json`. Source: `apps/cli/src/lib/fleet/remote-login.ts`, `apps/cli/src/lib/fleet/auth-sync.ts` (`FLEET_LOGIN_FLOWS`), `apps/cli/src/commands/ssh.ts`, `apps/cli/src/lib/open-url.ts`.
+
+- **`agents fleet ping` stops crying wolf on healthy accounts.** The auth matrix
+  painted a fully-logged-in fleet as half-broken: `codex`/`grok` (which have no
+  in-repo live-probe endpoint) rolled up as an alarming yellow `0/N`, and the
+  `--verbose` per-account list painted `expired` **red** — lumped with a real
+  `revoked` — even though `expired` is soft and self-refreshes on the CLI's next
+  launch (kimi/droid). Both renderers now share one truthful color model
+  (`verdictColor` / `authCellColor`): red is reserved for `revoked` (the only
+  "re-login now"); `unverified` reads as neutral **gray** "signed in
+  (unverifiable)"; `expired`/`rate_limited`/`error` are soft **yellow**; and the
+  cell numerator counts signed-in accounts (`live + present`) so a logged-in codex
+  fleet reads `1/1`, not `0/1`. Separately, `fleet ping --verbose` now actually
+  emits the per-account breakdown: the root program's global `--verbose` was
+  shadowing the subcommand flag, so the breakdown was silently unreachable — the
+  action now reads the effective value from the merged globals. Source:
+  `apps/cli/src/lib/auth-health.ts`, `apps/cli/src/commands/ssh.ts`,
+  `apps/cli/src/lib/auth-health.test.ts`.
+
+- **Wire allowlist support for ForgeCode and Hermes (RUSH-1748, RUSH-1749).** ForgeCode now receives permission groups as `~/.forge/permissions.yaml` operation-family policies (`read`, `write`, `command`, `url`) for built-in tools; this file is active only when `.forge.toml` has `restricted = true`, and MCP tools bypass it. Hermes now receives command allow rules in `~/.hermes/config.yaml` `command_allowlist` and deny globs in `approvals.deny`, preserving sibling YAML keys such as `mcp_servers` and `hooks`. Source: `apps/cli/src/lib/agents.ts`, `apps/cli/src/lib/permissions.ts`, `apps/cli/src/lib/resources/permissions.ts`, `apps/cli/src/lib/staleness/detectors/permissions.ts`, `apps/cli/docs/{00-concepts,02-resource-sync}.md`.
+
+- **`agents run --lease`/`--box` fail fast outside a git repo instead of billing a dead box.** crabbox syncs the working directory to the leased box with `git ls-files`, so from a non-git directory the run died at `build sync file list: exit status 128` — but only *after* provisioning (and billing) the box. `agents run` now checks the working directory is a git repo before provisioning and exits with an actionable message (`… is not a git repository. Run from inside a git repo, or initialize one: (cd <dir> && git init)`). Source: `apps/cli/src/commands/exec.ts`.
+
+- **Fix `agents run --lease` setup-copy (and `agents ssh <slug>`) to actually reach the box.** Both used a raw `ssh crabbox@ip:2222`, which fails `publickey` — crabbox provisions a per-lease identity key. They now tunnel through crabbox's own ssh invocation (`crabbox ssh --id <slug> --reclaim`), so the git-tracked `~/.agents` config really lands on a leased box (`agents repo refresh` then materializes it) and `agents ssh <slug>` connects. Verified end-to-end on a real Hetzner box. Source: `apps/cli/src/lib/crabbox/{setup-copy,cli}.ts`, `apps/cli/src/commands/ssh.ts`.
+
+- **Fix the Linear `--label` trigger filter matching nothing.** `routines` and
+  `monitors` jobs triggered on `linear:Issue` with `--label <name>` never fired:
+  the matcher read `data.labels.nodes[].name` (the GraphQL-query connection
+  shape), but Linear webhook bodies flatten list relations — `data.labels` is a
+  flat array of label objects. The `.nodes` read always yielded `[]`, so every
+  label filter silently failed to match. It now reads the flat array. The prior
+  unit test fixtured the same wrong shape, so the suite was green while the
+  integration was dead; the fixture now uses the real webhook shape and a
+  regression test locks it. Source: `apps/cli/src/lib/triggers/webhook.ts`.
+
+- **`agents menubar` now works from the Bun single-file binary.** When the CLI runs
+  as the compiled Bun executable, `import.meta.url` points inside the virtual
+  `/$bunfs/` bundle, so the menu-bar helper couldn't find the shipped
+  `MenubarHelper.app` on disk (`bundle source: missing (cannot enable)`) or read its
+  own `package.json` (`current version: unknown`, and a perpetual "stale" warning).
+  `enable` refused with "no menu-bar helper bundle ships with this install." Version
+  and bundle resolution now fall back to the real on-disk install, located by
+  following the `agents` launcher symlink, so `enable`/`disable`/`status` behave the
+  same whether the CLI runs under Node or the Bun binary. Source:
+  `apps/cli/src/lib/version.ts`, `apps/cli/src/lib/menubar/install-menubar.ts`.
+
+- Add OpenClaw workflow sync by projecting agents-cli workflows into Lobster `.lobster` files under `.openclaw/workflows/`.
+
+- **Fix the PTY sidecar (`agents pty`, interactive `agents teams`, `agents fleet login`) on the macOS standalone binary.** Since the macOS release became a `bun --compile` standalone (#315), the sidecar was spawned AS that binary (`process.execPath pty _server`) — but a Bun standalone cannot `require()` a native addon, so node-pty's `pty.node` failed to load (`Cannot require module ../build/Debug/pty.node`) and every PTY-backed command died with "PTY server failed to start within 5 seconds." `getServerSpawnArgs` now detects the standalone case and runs the sidecar via a real `node` executing the `dist/index.js` that ships beside the binary (where the prebuilt `pty.node` loads from disk), falling back to the binary only when no node / no dist is found. Verified end-to-end against a real compiled Mach-O standalone. Source: `apps/cli/src/lib/pty-client.ts`.
+
+- `agents sessions` now indexes routine-run transcripts from durable run history; use `agents sessions --routine --all` or `agents sessions <run-id>` to inspect a routine run with the existing summary view.
+
+- **`agents routines add`, `run`, and `runs` now support `--json`.** Previously only
+  `list`/`status` emitted JSON, so an agent creating a routine or triggering a run
+  had to scrape human strings for the job name / run id. `add` emits
+  `{ ok, added, job }`, `run` emits `{ ok, job, runId, logDir }`, and `runs` emits an
+  array of run records — all on stdout, with the scheduler-start banner suppressed so
+  it never pollutes the JSON stream. Source: `apps/cli/src/commands/routines.ts`.
+  (RUSH-1833)
+
+- **`agents run <agent>` no longer hangs when launched headless without a prompt.**
+  A run with no prompt and no explicit `--interactive` resolves to interactive
+  intent — but in a non-TTY shell (a headless agent, a pipe, CI) there is no
+  terminal to host the REPL, so it attached a TUI to dead stdin and hung forever.
+  It now fails fast with the headless alternatives (`agents run <agent> "<task>"`
+  or `agents run <agent> --headless` to read the prompt from stdin). An explicit
+  `--interactive` is still honored. Source: `apps/cli/src/commands/exec.ts`,
+  `apps/cli/src/lib/exec.ts` (`inferredInteractiveWithoutTty`). (RUSH-1829)
+
+- Menu bar routines now include latest run `exitCode` and `failureReason` from `agents routines list --json`, show failed routine reasons inline, label healthy-but-overdue routines as `overdue` instead of `exit 0`, and open the concise logs summary instead of a raw Terminal dump.
+
+- Show Droid teammate activity in `agents teams collect` by normalizing stream-json tools, file edits, and final messages.
+
+Fix: route remaining specialized direct SSH spawns through the shared hardened SSH baseline.
+
+- **Agent feed dispatch now keeps local and remote answer paths separate (RUSH-1472).** `agents feed` only applies stall suppression, default-on-no-answer policy, and dispatch controls to blocks owned by the local machine, so remote feed rows cannot enqueue answers into the wrong local mailbox. Per-block `timeoutMinutes` is honored for approval defaults and decision parking, policy/default answers are tested against the real mailbox spool, block-specific `allowedOperators` restrict high-consequence answers, and urgent notification text is emoji-free. Source: `apps/cli/src/commands/feed.ts`, `apps/cli/src/lib/{ask-classifier,feed,feed-policy,notify}.ts`.
+
+- **`agents feed` ranks blocked agents by cost of delay and surfaces runaway/needy control cards (RUSH-1478).** Open blocks are sorted by idle time, downstream blast radius, dollar burn rate, and classifier irreducibility, while silent high-burn/relaunch-loop agents and chronic askers render once as control cards with `ag feed --pause` / `ag feed --kill` actions. Source: `apps/cli/src/lib/feed-ranking.ts`, `apps/cli/src/commands/feed.ts`, `apps/cli/src/lib/feed.ts`.
+
+- Make live session follows compact by default: `agents sessions tail` now prints low-noise message/tool/result lines unless `--json` is passed, and `agents logs -f <id>` keeps raw transcript streaming behind `--full`.
+
+- **Fix `agents repo refresh` for stale plugin skill shadows.** Full refresh now forces a materialization pass, copies trusted plugin-bundled skills into legacy top-level agent-home skill dirs when those names already matter to the agent, and prunes orphaned top-level skill dirs whose source no longer exists.
+
+- `agents run <profile> --lease` now provisions the profile's host runtime and temporary profile config on the leased box without copying base-runtime OAuth credentials when the profile authenticates with its own API key, including OpenCode and Antigravity-hosted profiles.
+
+- `agents pty` now starts its sidecar correctly from the standalone CLI binary and includes the spawned command plus recent sidecar log lines when startup fails.
+
+- Fixed `agents secrets openclaw-keychain migrate` so OpenClaw Keychain writes no longer expose secret values in process argv, and migration now fails closed when matching plaintext credentials disagree.
+
+- `agents share setup` and `agents setup share` now read Cloudflare provisioning credentials from the `cloudflare` secrets bundle and persist the Worker write token as `WRITE_TOKEN` in the `share` bundle, matching the setup/onboarding contract while keeping endpoint config in `agents.yaml` under `share:`.
+
+- Fix `agents routines list` and `agents routines view` so a project-layer routine with the same name no longer hides the user-layer `devices` allowlist written by `agents routines devices --set`.
+
+- Keep piped CLI output human-readable unless `--json` is passed.
+
+- Add `--json` to `agents monitors view` and `agents monitors test`, and send monitor errors to stderr so JSON stdout stays parseable.
+
+- Add `--json` output to `agents routines add`, `agents routines run`, and
+  `agents routines runs` so scripts can capture routine and run ids without scraping human text.
+
+- Fix hook directory sync status so bundled hook directories such as `hooks/tests` copy correctly and no longer appear permanently drifted after sync.
+
+- Neutralize residual OSS scrub breadcrumbs by replacing browser/session test fixture hostnames with `remote-host` and removing legacy private product path references from cloud proxy comments.
+
+- **Add `agents setup mine` / `agents mine` — white-label the CLI.** Mint your own personally-named binary (e.g. `jack`) that runs every agents verb under your name, with the built-in commands you disable hidden and a per-brand resource profile that curates skills/plugins/MCP/etc. `agents setup mine` is the wizard; `agents mine init/list/toggle/remove` manage brands. Free and Apache-2.0. Source: `apps/cli/src/commands/mine.ts`, `apps/cli/src/lib/brand.ts`.
+
+- **One-time "star us on GitHub" nudge after your first successful run.** After a
+  user's first successful `agents run` or `agents teams`, agents-cli prints a
+  single plain inline line pointing at the repo. Shown at most once ever (claimed
+  with an atomic O_EXCL sentinel so concurrent `agents teams` processes can't
+  double-print), and skipped for non-TTY, CI, `--json`/`--quiet`, or
+  `AGENTS_NO_NUDGE=1`. The `agents teams` call site only nudges on a clean drain
+  (no failed teammates). Source: `apps/cli/src/lib/star-nudge.ts`,
+  `apps/cli/src/commands/exec.ts`, `apps/cli/src/commands/teams.ts`,
+  `apps/cli/src/lib/teams/supervisor.ts`.
+
 ## 1.20.72
 
 - **Stop `agents doctor` from reporting phantom drift and `agents prune` from

package/README.md

@@ -35,12 +35,16 @@
 
 https://agents-cli.sh/demo.mp4
 
+## Quickstart
+
 ```bash
-npm install -g @phnx-labs/agents-cli
-# or
-bun install -g @phnx-labs/agents-cli
+npm install -g @phnx-labs/agents-cli   # or: curl -fsSL agi-cli.sh | sh
+agents setup                           # first-time setup -- config + pick your agents
+agents run claude "explain this repo"  # run any agent on your existing subscription
 ```
 
+`agents setup` is interactive and idempotent -- safe to re-run on a new machine. The `agi-cli.sh` one-liner installs this same canonical `@phnx-labs/agents-cli` package. Prefer bun? `bun install -g @phnx-labs/agents-cli` works too.
+
 Already installed? `agents upgrade` updates agents-cli itself to the latest version (`agents upgrade 1.2.3` for a specific version or dist-tag, `-y` to skip the confirm prompt). The command is `upgrade` on every platform -- there is no `agents update` (on macOS, `agents helper update` is a different command that reinstalls the keychain helper, not agents-cli).
 
 Source: [github.com/phnx-labs/agents-cli](https://github.com/phnx-labs/agents-cli)
@@ -604,6 +608,7 @@ Bundle skills, commands, hooks, MCP servers, settings, and permissions under a s
 # Install from a git URL or local path
 agents plugins install hivemind@https://github.com/activeloopai/hivemind.git
 agents plugins install ./my-plugin
+agents plugins add ./my-plugin
 
 # Apply to one agent (default version) or all supported
 agents plugins sync rush-toolkit claude
@@ -646,6 +651,32 @@ Plugins live in the user repo (`~/.agents/plugins/`), not inside any single vers
 
 ---
 
+## Make it yours
+
+White-label the CLI. `agents setup mine` mints a **personally-named binary** — `jack` instead of `agents` — that _is_ agents-cli: same tool, your name, running the exact feature set you choose. Anyone can mint their own; Jack and Pranjal each get an independent brand.
+
+```bash
+agents setup mine                      # wizard: pick a name, check off what to disable
+agents mine init jack --disable teams cloud   # or non-interactively
+
+jack run claude "hello"                # every agents verb, under your name
+jack --help                            # help, version, and errors all read "jack"
+```
+
+Manage brands with `agents mine list | toggle | remove`:
+
+```bash
+agents mine toggle jack --disable-plugin rush --disable-skill deploy
+agents mine toggle jack --enable teams
+agents mine remove jack --purge
+```
+
+Under the hood, `init` drops a pure pass-through shim in `~/.agents/.cache/shims/<name>` (already on `PATH`) that sets `AGENTS_BRAND` and forwards every argument to the same binary — nothing is copied or forked. The brand's config lives in `~/.agents/agents.yaml` (`brands.<name>`), so it rides `agents repo push/pull` across your fleet. Disabling a command hides it **only** under that brand; plain `agents` / `ag` keep every command. Curated skills/plugins/MCP ride a per-brand [resource profile](apps/cli/docs/profiles.md). Full reference: [Make it yours](apps/cli/docs/mine.md).
+
+> Personal use is free and Apache-2.0. Redistributing a branded build commercially will require a license in a future release.
+
+---
+
 ## Browser
 
 <p align="center">
@@ -862,6 +893,7 @@ Sources: a command's stdout (`--watch` / `--poll`), an HTTP endpoint (`--poll-ht
 # Publish an HTML artifact to a public link on your own Cloudflare R2 (~$0).
 agents share setup                                  # once: provision bucket + Worker on your CF
 agents share plan.html --slug fleet --expire 30d    # → https://<base>/fleet
+agents share plan.html --json                       # URL object for plan-render hooks
 agents share status                                 # show the endpoint
 ```
 
@@ -877,6 +909,8 @@ this is effectively free.
 **Fleet mode:** provision one endpoint, then every fleet / cloud / ephemeral agent
 publishes through it with a shared write token — `agents share join <baseUrl>` uses an
 existing endpoint with no provisioning. `--expire 30d|12h|<date>` auto-expires a link.
+`--json` emits `{ url, coverUrl, expiresAt }` so plan-render automation can publish the
+rendered HTML and post the returned link without scraping terminal text.
 See [docs/share.md](apps/cli/docs/share.md).
 
 ---
@@ -924,7 +958,7 @@ Two repos with the same shape, different roles:
 
 **Version pinning:** `agents.yaml` at project root pins which agent version to use (like `.nvmrc` for Node).
 
-**Resource resolution:** When syncing resources (commands, skills, rules, hooks, MCP, permissions), the order is **project > user > system**. A `.agents/` directory at project root wins, then `~/.agents/`, then `~/.agents-system/`. Same-named resources higher in the chain override lower ones; everything else unions in.
+**Resource resolution:** When syncing resources (commands, skills, rules, hooks, MCP, permissions), the order is **project > user > system**. A `.agents/` directory at project root wins, then `~/.agents/`, then `~/.agents-system/`. Same-named resources higher in the chain override lower ones; everything else unions in. Run `agents resources --merged` to see the effective skills, commands, MCP servers, hooks, rules, plugins, workflows, and subagents, with each row tagged by its winning layer.
 
 See [docs/00-concepts.md](apps/cli/docs/00-concepts.md) for the full mental model: DotAgents repos, resource kinds, and how resolution works end-to-end.
 
@@ -1151,6 +1185,8 @@ Yes -- `agents run` is non-interactive by default. `--yes` auto-accepts prompts,
 
 The auto-update prompt is suppressed automatically when stdin or stdout isn't a TTY. For headless environments where TTY detection misfires (k8s pods that allocate a PTY for stdout, cloud sandbox factories), set `AGENTS_CLI_DISABLE_AUTO_UPDATE=1` to skip the update check entirely -- no prompt, no network call.
 
+agents-cli also prints a one-time "star us on GitHub" line after your first successful `agents run`/`agents teams`. It's already skipped in CI, non-TTY, `--json`, and `--quiet` runs; set `AGENTS_NO_NUDGE=1` to suppress it everywhere.
+
 To update on demand instead of waiting for the prompt, run `agents upgrade` (add `-y` to skip the confirmation, or pass a version/dist-tag to install something other than latest).
 
 ### What happens to my config when I switch versions?

package/dist/commands/activity.d.ts

@@ -0,0 +1,12 @@
+/**
+ * `agents activity` -- the agent-activity lane: a running stream of what agents
+ * DID (plan created, PR opened, worktree created, sub-agent spawned, files
+ * edited), read from the append-only per-session activity logs.
+ *
+ * Complements `agents feed` (what agents are WAITING on). Events are emitted at
+ * hook time by 11-activity-log.py, so this reader never re-parses transcripts --
+ * it tails the logs and collapses routine work to counts, surfacing milestones
+ * individually and newest-first.
+ */
+import type { Command } from 'commander';
+export declare function registerActivityCommand(program: Command): void;

package/dist/commands/activity.js

@@ -0,0 +1,68 @@
+import chalk from 'chalk';
+import { collapseActivity, ensureActivityLogHook, formatActivityLine, readRecentActivity, styleForEvent, tierForEvent, } from '../lib/activity.js';
+/** Wire the activity-log hooks into each hooks-capable version's settings. */
+async function installActivityHooks() {
+    const warnings = [];
+    const hookInstall = ensureActivityLogHook();
+    if (hookInstall.error) {
+        warnings.push(hookInstall.error);
+        return warnings;
+    }
+    const [{ iterHooksCapableVersions, parseHookManifest, registerHooksToSettings }, { getVersionHomePath }] = await Promise.all([
+        import('../lib/hooks.js'),
+        import('../lib/versions.js'),
+    ]);
+    const manifest = parseHookManifest({ warn: false });
+    for (const { agent, version } of iterHooksCapableVersions({ agent: 'claude' })) {
+        const result = registerHooksToSettings(agent, getVersionHomePath(agent, version), manifest);
+        if (result.errors.length > 0)
+            warnings.push(`${agent}@${version}: ${result.errors.join('; ')}`);
+    }
+    return warnings;
+}
+export function registerActivityCommand(program) {
+    program
+        .command('activity')
+        .description('Recent agent activity -- plans, PRs, worktrees, sub-agents (newest first)')
+        .option('--json', 'Emit the raw event list as JSON')
+        .option('--all', 'Include routine activity (file edits) inline, not collapsed')
+        .option('--milestones', 'Only milestone events (plans, PRs, worktrees, sub-agents)')
+        .option('-n, --limit <n>', 'Cap the number of events shown', (v) => parseInt(v, 10), 40)
+        .option('--since <minutes>', 'Only events within the last N minutes', (v) => parseInt(v, 10))
+        .action(async (opts) => {
+        const warnings = await installActivityHooks();
+        const sinceMs = typeof opts.since === 'number' && opts.since > 0
+            ? Date.now() - opts.since * 60_000
+            : undefined;
+        let events = readRecentActivity({ sinceMs, limit: opts.limit });
+        if (opts.milestones)
+            events = events.filter((e) => tierForEvent(e.event) === 'milestone');
+        if (opts.json) {
+            process.stdout.write(`${JSON.stringify(events, null, 2)}\n`);
+            return;
+        }
+        for (const w of warnings)
+            process.stderr.write(chalk.yellow(`  ! ${w}\n`));
+        if (events.length === 0) {
+            process.stdout.write(chalk.gray('No recent agent activity. Events appear here as agents create plans, open PRs, and spawn sub-agents.\n'));
+            return;
+        }
+        const { milestones, counts, subagentCount } = collapseActivity(events);
+        process.stdout.write(chalk.bold('\n  recent activity\n'));
+        const shown = opts.all ? events : milestones;
+        for (const ev of shown)
+            process.stdout.write(`${formatActivityLine(ev, { showHost: true })}\n`);
+        if (!opts.all) {
+            const parts = Object.entries(counts)
+                .sort((a, b) => b[1] - a[1])
+                .map(([event, n]) => `${styleForEvent(event).label} ×${n}`);
+            if (parts.length > 0) {
+                process.stdout.write(chalk.gray(`\n  · ${parts.join('  ')}\n`));
+            }
+        }
+        if (subagentCount > 0) {
+            process.stdout.write(chalk.magenta(`\n  ⑂ ${subagentCount} sub-agent${subagentCount === 1 ? '' : 's'} spawned\n`));
+        }
+        process.stdout.write('\n');
+    });
+}

package/dist/commands/apply.d.ts

@@ -9,4 +9,17 @@
 import { Command } from 'commander';
 /** padEnd on the visible width, ignoring chalk color codes. Exported for tests. */
 export declare function stripPad(s: string, width: number): string;
+/**
+ * Attach the reconcile options + action to a command node. Shared by the
+ * top-level `agents apply` and its `agents fleet apply` alias so the two can
+ * never drift — identical flags, identical engine.
+ */
+export declare function configureApplyCommand(cmd: Command): Command;
 export declare function registerApplyCommand(program: Command): void;
+/**
+ * Surface `agents apply` under the fleet command tree as `agents fleet apply`
+ * (and `agents devices apply`). Same reconcile engine as the top-level command —
+ * a discoverability alias for users who reach for `fleet` as the verb. Kept in
+ * lockstep via {@link configureApplyCommand}.
+ */
+export declare function registerFleetApplyAlias(devicesCmd: Command): void;