@lmctl-ai/lmctl 0.1.85 → 0.1.87

package/CHANGELOG.md

@@ -2,6 +2,48 @@
 
 All notable public-preview changes for `@lmctl-ai/lmctl` are recorded here.
 
+## 0.1.87 — 2026-07-07
+
+- **`lmctl wait`/`lmctl exec` are now the synchronous wake model, on a new
+  `tracked_invocation` substrate.** Every backgrounded lmctl invocation records
+  a start row and a finish row; `lmctl wait` blocks on those rows and returns the
+  instant the first one in scope finishes (reporting which completed and which
+  are still running). `lmctl exec -- <command>` runs any shell command as a
+  tracked invocation, and a backgrounded `lmctl chat … --from "<teamfile>:Lead"`
+  is tracked the same way — so `cmd & lmctl wait` is the wake for chats and
+  arbitrary commands alike. This replaces the fragile "reserve one blocking job
+  as the wake" (N−1/1) pattern; all N launches go to the background and `wait` is
+  the single return-point.
+- **Removed `lmctl chat --detach`, `lmctl jobs`, `lmctl nudge`, and turn-exit
+  auto-delivery.** The old detached-worker / `delegation_job` background model is
+  gone, along with its per-provider (codex/gemini) poll special-casing — every
+  provider now uses the same background-then-`wait` loop. The legacy
+  `delegation_job` / `delegation_event` tables are **kept dormant** for existing
+  DB compatibility (no destructive migration); only the active code paths were
+  removed.
+- **`wait` is robust against the failure modes of a fire-and-forget loop.** A
+  phantom `running` row whose process died without writing a finish is reclaimed
+  as `error` (dead/stale holder detection) instead of hanging `wait` forever; an
+  appearance-grace poll covers the `cmd & lmctl wait` insert-lag race; and
+  `lmctl exec --json` prints its invocation id so `lmctl wait --id <id>` gives a
+  zero-race handshake. Scopes: default (caller, via `LMCTL_SELF_SESSIONID`),
+  `--from`, `<teamfile>`, `--all`, `--id`; exit `0` completed/idle, `1` timeout
+  (8h backstop), `2` usage. A single `idle` return means "nothing running right
+  now", not "done" — loop `wait`.
+
+## 0.1.86 — 2026-07-07
+
+- **New `lmctl wait` — the wake primitive for the meta-lead orchestration loop.**
+  Dispatch N tracked background jobs (`lmctl chat --detach`), then `lmctl wait`
+  blocks (polling `delegation_job`) and returns the moment the first snapshot job
+  finishes — reporting which completed and which of that snapshot are still in
+  flight. Defaults to the caller's own dispatched jobs (`--all` widens, a
+  `<teamfile>` positional scopes); returns `idle` when nothing is in flight (so
+  the loop knows to pull fresh work from its queue); no default timeout
+  (`--timeout` optional); polls every 3s (`--interval`). It burns no tokens while
+  blocked and works across all providers. Replaces the fragile "reserve one
+  blocking job as the wake" (N-1/1) pattern.
+
 ## 0.1.85 — 2026-07-07
 
 - **lmplayer: a delegation job that completes with empty output is no longer

package/README.md

@@ -49,16 +49,17 @@ lmctl chat ./my-project/my-project.lmctl Lead "Inspect the repo and propose next
 ```
 
 Start with `lmctl run <url>` for published workflows. For a custom AI-Lead team,
-use the plan -> lint -> seed -> chat flow. Use provider-native background
-execution for long turns, with `lmctl jobs` as the tracked work portal. Daemon,
-web, API, and MCP surfaces are secondary integration paths; the CLI/job portal
-is the golden path for onboarding.
+use the plan -> lint -> seed -> chat flow. For long turns, background a tracked
+invocation (`lmctl exec -- <command> &` or a backgrounded `lmctl chat`) and let
+`lmctl wait` block until it finishes — that return is the wake. Daemon, web, API,
+and MCP surfaces are secondary integration paths; the CLI is the golden path for
+onboarding.
 Lintable teamfile templates for research, writing, and review teams ship in the
 [examples gallery](https://lmctl.com/examples/).
 
-`lmctl` top-level subcommands are: `api`, `chat`, `code`, `serve`, `intake`,
-`mcp`, `monitor`, `device`, `team`, `terminal`, `tail`, `health`, `jobs`,
-`nudge`, `recover`, `ls`, `lint`, `seed`, `hire`, `refresh`, `clone`, `plan`, `workflow`,
+`lmctl` top-level subcommands are: `api`, `chat`, `loop`, `code`, `serve`, `intake`,
+`mcp`, `monitor`, `device`, `team`, `terminal`, `tail`, `health`, `wait`,
+`exec`, `recover`, `ls`, `lint`, `seed`, `hire`, `refresh`, `clone`, `plan`, `workflow`,
 `run`, `project`, `workspace`, `diagnose`, `diagnose-prompt`, `status`, and
 `db`. Run `lmctl <command> --help` for details on each, and see the
 [documentation](https://lmctl.com/lmctl) to define your first team.