@lmctl-ai/lmctl 0.1.83 → 0.1.85

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -2,6 +2,57 @@
2
2
 
3
3
  All notable public-preview changes for `@lmctl-ai/lmctl` are recorded here.
4
4
 
5
+ ## 0.1.85 — 2026-07-07
6
+
7
+ - **lmplayer: a delegation job that completes with empty output is no longer
8
+ silently marked `done`.** A detached `provider=lmplayer` job whose reply is
9
+ empty is now corroborated against the session store — if the session is missing
10
+ or has zero persisted messages, the job is terminalized as `error`
11
+ (`empty_lmplayer_result`) instead of a silent success, so the lost work is
12
+ visible instead of masquerading as done. Non-empty replies and non-lmplayer
13
+ jobs are unaffected.
14
+
15
+ ## 0.1.84 — 2026-07-07
16
+
17
+ - **lmplayer: `tail`/`health`/`ls` now find sessions on a fresh install.**
18
+ lmplayer writes a channel-suffixed session DB (`opencode-<channel>.db`, e.g.
19
+ `opencode-dev.db`) rather than `opencode-local.db`; the reader now scans all
20
+ lmplayer/opencode-compatible DB candidates and reads from the one that actually
21
+ contains the session (skipping stale/incompatible stores).
22
+ - **Fixed the `model_selection.md not found` warning on global installs.** The
23
+ tested-models catalog is now shipped in the package, and the runtime resolver
24
+ finds it across source, package-root, and bundled-`dist` layouts; a
25
+ `provider=lmplayer model=…` member no longer emits a spurious "cannot be
26
+ verified" warning.
27
+
28
+ ## 0.1.83 — 2026-07-06
29
+
30
+ - **`lmctl terminal` takes a DB lock** (host + pid + heartbeat) so a second chat
31
+ to a member held live in a terminal is rejected instead of racing; a stale/dead
32
+ or remote lock is reclaimed safely, and `--force` overrides.
33
+ - **`--detach` is hidden from the CLI surface.** It stays functional for AI-team
34
+ chat but is the codex/gemini-only fallback; other providers use their native
35
+ tracked-background path.
36
+ - **MCP install retired.** lmctl no longer writes provider MCP config on
37
+ seed/chat (the internal `lmctl_chat` code remains; only the installation is
38
+ removed).
39
+ - **Skills over the network.** The default teamfile template pulls shared skills
40
+ (background-wakeup, team-lead-workflow, durable-memory) from lmctl.com via
41
+ `_SKILL_ <url>` into `durable-memory/skills/` on seed.
42
+ - **`effort=` advisory for `provider=qwen`.** Qwen Code exposes no reasoning-effort
43
+ CLI flag, so lint points you to a thinking/effort model id instead of
44
+ mis-warning that effort is ignored.
45
+
46
+ ## 0.1.82 — 2026-07-06
47
+
48
+ - **New first-class provider: `provider=lmplayer`.** A one-shot `lmplayer run`
49
+ adapter — seed captures the sessionID from the JSON stream, resume via
50
+ `run --session`, `effort=`, YOLO via `--dangerously-skip-permissions`, and it
51
+ reuses the opencode session-reader against lmplayer's store so
52
+ `tail`/`health`/`ls` work.
53
+ - **`lmctl loop`** — a simple autopilot driver that repeatedly continues a member
54
+ until it reports `ALL DONE` or `OPERATOR ESCALATION`.
55
+
5
56
  ## Unreleased
6
57
 
7
58
  - Teamfile templates no longer hardcode `sessiondir`: `plan`/`hire` omit it