@loompad/cli 1.2.3 → 1.2.5

package/docs/release-notes/v1.2.4.md

@@ -0,0 +1,13 @@
+## Loom 1.2.4: two machines, one goal, set up in one step
+
+### Set up team auto-orchestration in one step
+- **`loom team auto`** checks whether goals can split across your teammates' machines here. It says which of their agents a goal would get, and what's missing.
+- **`loom team auto on`** does the setup:
+  - shares the project;
+  - writes `loom.team.json` if the repo has none (commit and merge it);
+  - turns on your machine's shared runner, so teammates' goals can use your agents too.
+- **The Team tab** shows the same status on a **Two machines, one goal** card, with a **Turn on** button.
+
+### Steadier on a teammate's machine
+- **A teammate's task no longer fails to commit.** Loom on their machine was saving its own end-of-turn checkpoint at the same moment and could undo the task's changes. The task's commit is now built separately, so nothing interferes.
+- **Follow-ups reach the teammate.** When your orchestrator asks a teammate's task for more ("also fix the test"), their machine now gets that request on top of its earlier work, instead of the whole task again.

package/docs/release-notes/v1.2.5.md

@@ -0,0 +1,19 @@
+## Loom 1.2.5: steadier when agents do real work
+
+### Updates don't kill running work
+- **A newer CLI leaves a busy daemon alone.** It used to restart a daemon on an older build automatically, killing every turn in flight. Now it says to restart once the turns finish. `loom up --restart` refuses while turns run, unless you add `--force`.
+- **`send --wait` says when a turn was interrupted.** It exits 75 with the command to carry on, instead of "fetch failed".
+
+### Every agent starts, or says why
+- **opencode starts with `OPENCODE_SERVER_PASSWORD` set.** Loom gives the server it starts a password of its own.
+- **An opencode session that silently drops prompts no longer hangs.** Within 45 seconds Loom retries in a fresh session, or names the model that won't run.
+- **A crashed opencode server ends the turn at once** instead of leaving the agent "busy" for up to an hour.
+- **antigravity finishes its builds.** Every turn tells it the turn ends at its reply, so it runs tests in the foreground instead of waiting for a notification that never comes.
+
+### Agent access
+- **`loom agents:mode <agent> [bypass|auto|ask]`** shows what each mode means and switches it. For example, codex's default sandbox has no network and a read-only `.git`. `loom agents` shows each agent's access.
+
+### Honest scripts
+- `send` to an agent without the baton fails in a script instead of quietly exiting 0. `--handoff` takes the baton.
+- A turn that reports no cost says `null`, not `$0`.
+- `loom agents` outside a project lists what this machine can run. `loom status` marks the project you're in. `loom init` in a script keeps `.loom/` out of git.

package/docs/teams.md

@@ -171,6 +171,9 @@ machine. Your orchestrator sees each teammate's agents as `team:<agent>` workers
 self-contained tasks go to a teammate, tightly coupled ones stay with you. Nobody
 on their side clicks anything.
 
+`loom team auto` checks it's ready and says what's missing; `loom team auto on` does the
+setup below for you (the Team tab's **Turn on** is the same thing).
+
 ```bash
 # once, in the repo: let teammates' runners take its work (reviewed like any policy change)
 echo '{ "runners": { "shared": true } }' > loom.team.json

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@loompad/cli",
-  "version": "1.2.3",
+  "version": "1.2.5",
   "description": "Loom — local-first control plane for coding agents: one CLI, shared memory, baton handoffs and routes across Claude Code, OpenCode, and friends.",
   "type": "module",
   "license": "MIT",