@gapi/gcli 1.8.239 → 1.8.241

package/docs/changelog.d/2026-10-19-dedicated-platform-cluster-cluster-scaling-topology-load-balancer-and-health-commands-c.md

@@ -0,0 +1,31 @@
+## 2026-10-19 - dedicated-platform-cluster: Cluster scaling, topology, load balancer and health commands (cluster:status, health, scale, autoscale, ...)
+
+**Feature:** dedicated-platform-cluster · **Type:** feat
+
+- New commands (the dashboard cluster page's actions, same
+  cluster-provisioner operations): `cluster:status`, `cluster:health`,
+  `cluster:scale-events`, `cluster:cost`, `cluster:scale --workers`,
+  `cluster:scale:control-planes --count`, `cluster:multi-node`,
+  `cluster:single-node`, `cluster:autoscale` (partial: only the given
+  flags change), `cluster:load-balancer --enable|--disable`. All take
+  `--cluster <id>` or `--clusterName <name>`; mutations take `--wait`
+  (polls until ready; ending in `error` exits non-zero with the reason).
+- `cluster:apply [--spec] [--dry-run] [--yes]`: the `cluster:` section of
+  lambforge.yaml / cluster.yaml is now a desired state (new keys
+  `controlPlanes`, `loadBalancer`, `autoscale.*`; `singleNode`, `workers`
+  as before). It creates the cluster when missing (in the desired shape)
+  and otherwise plans and applies topology → workers → control planes →
+  load balancer → autoscale, waiting for `ready` between jobs; destructive
+  steps need `--yes`. `cluster:create` also passes `controlPlanes` and
+  `loadBalancer` from the spec now.
+- Client-side checks mirror the server's bounds (workers 1-10, control
+  planes 1/3/5, autoscale ranges) so a bad value fails before any request.
+- `README.md` Private Clusters section, regenerated `README-CLI.md`.
+- **Test:** `npx jest --coverage=false` (71 tests; new
+  `test/cluster-scaling.spec.ts`, `test/cluster-apply.spec.ts` (the plan),
+  `test/cluster-apply-command.spec.ts` (dry-run, `--yes`, create, refusals)).
+- **Debug:** slice doc `docs/features/dedicated-platform-cluster.md`.
+- **Verification:** jest 71/71; the built binary's `cluster:apply` read a
+  real lambforge.yaml and rejected `controlPlanes: 2` before any request; tsc + eslint clean; `npm run build` and
+  `--help` checked for every command (`release/index.js` restored after,
+  not a release).

package/docs/changelog.d/2026-10-19-lambforge-agent-skill-readme-points-to-the-lambforge-agent-skill-gcli-mcp-start-fo.md

@@ -0,0 +1,17 @@
+## 2026-10-19 - lambforge-agent-skill: README points to the Lambforge agent skill (gcli + mcp:start for AI agents)
+
+**Feature:** lambforge-agent-skill · **Type:** docs
+
+- README.md "AI agent skill" (under AI Integration): the prompt to paste into
+  an agent (https://lambforge.com/skills/platform.md), the manual Claude Code
+  install (every file from `files.txt`, `platform.md` copied to `SKILL.md`,
+  `claude mcp add --scope user lambforge -- gcli mcp:start`), and what the
+  topic files cover. The skill lives in gql-fission `src/skills/` (canonical
+  doc there).
+- `docs/features/lambforge-agent-skill.md`: which gcli commands the skill
+  quotes - change them, update SKILL.md in the same change.
+- **Test:** none (docs); the commands it lists were checked with
+  `node release/index.js <command> --help`.
+- **Debug:** an agent runs a gcli command that fails → compare SKILL.md with
+  `gcli <command> --help`.
+- **Verification:** no code change.

package/docs/changelog.d/2026-10-19-repo-cron-create-honours-cron-triggername-readme-brought-up-to-da.md

@@ -0,0 +1,21 @@
+## 2026-10-19 - repo: cron:create honours --cron/--triggerName; README brought up to date with every command
+
+**Feature:** repo · **Type:** fix
+
+- `cron:create` declared `--cron` and `--triggerName` but only read them
+  from the spec (`triggers.time` of lambforge.yaml / cron.yaml), so the
+  documented `gcli cron:create --lambdaName x --triggerName nightly --cron
+  '0 2 * * *'` sent an empty schedule. Flags now win, the spec fills gaps,
+  and a missing schedule is a clear error before any request.
+- README: binary install from `releases/latest` (was pinned to v1.8.198);
+  `environment:update/delete`, `lambda:list` and a "Schedules (cron)"
+  section (the commands were missing); `triggers:` in the unified
+  lambforge.yaml; the MCP client URL is `https://<host>/<projectId>/<route>`
+  (the `/mcp` suffix was wrong - see example/mcp/lambforge.yaml); the
+  declarative cluster section. Every command registered in `src/commands`
+  now appears in README.md.
+- **Test:** `npx jest --coverage=false test/cron-create.spec.ts` (flags,
+  flags over yaml, missing schedule).
+- **Debug:** a cron with an empty schedule → an older gcli; upgrade.
+- **Verification:** jest 71/71; a script compared the commands registered in
+  `src/commands` against README.md (no command missing).

package/docs/features/README.md

@@ -31,3 +31,5 @@ Outline (drop sections that don't apply):
 | Feature | Doc | Notes |
 |---|---|---|
 | `local-ai-runners` | [local-ai-runners.md](local-ai-runners.md) | `runner:*` commands; canonical doc in graphql-server |
+| `lambforge-agent-skill` | [lambforge-agent-skill.md](lambforge-agent-skill.md) | The agent skill that teaches gcli + `mcp:start`; canonical doc in gql-fission |
+| `dedicated-platform-cluster` | [dedicated-platform-cluster.md](dedicated-platform-cluster.md) | `cluster:status/health/scale/autoscale/...` commands; canonical doc in lambforge-agent |

package/docs/features/dedicated-platform-cluster.md

@@ -0,0 +1,70 @@
+# dedicated-platform-cluster (gcli slice)
+
+**Canonical doc:** `lambforge-agent/docs/features/dedicated-platform-cluster.md`
+· **Since:** 2026-10-19 (cluster scaling and health commands)
+
+The cluster page's scaling, topology, load balancer and health actions as
+commands (cluster-provisioner operations, the same ones gql-fission's
+`core/services/cluster/*.graphql` use; every one is USER-enabled with
+`isMemberOrProjectOwner`):
+
+| Command | Operation |
+|---|---|
+| `cluster:status [--cluster]` | `projectClustersOverview` (status, agent, nodes, load, autoscale, issues) |
+| `cluster:health --cluster` | `clusterHealth(limit: 1)` - latest agent report + nodes |
+| `cluster:scale-events --cluster [--limit]` | `clusterScaleEvents` |
+| `cluster:cost` | `projectClustersOverview` (`monthToDate`, `estimatedMonthlyCostEur`) |
+| `cluster:scale --workers 1-10 [--wait]` | `scaleProjectCluster` |
+| `cluster:scale:control-planes --count 1/3/5 [--wait]` | `scaleControlPlanesProjectCluster` |
+| `cluster:multi-node --workers [--wait]` / `cluster:single-node [--wait]` | `migrateToMultiNodeProjectCluster` / `revertToSingleNodeProjectCluster` |
+| `cluster:autoscale [--enable/--disable] [--min --max --cpuUp --memUp --cpuDown --memDown --samples --cooldown]` | `updateProjectClusterAutoscale` - partial: the server merges it into the stored policy and validates (min <= max, scale-down below scale-up, single-node cannot autoscale) |
+| `cluster:load-balancer --enable/--disable [--wait]` | `enable/disableLoadBalancerProjectCluster` |
+| `cluster:apply [--spec] [--dry-run] [--yes]` | reconciles the `cluster:` section of lambforge.yaml: `projectClusters` (current) → plan → provision / topology / scale / control planes / load balancer / autoscale, one job at a time, waiting for `ready` between them |
+
+## Declarative clusters (`cluster:apply`)
+
+`cluster:` keys: `name` (identity), `imageId` / `serverType` / `location`
+(create only), `singleNode`, `workers` (1-10), `controlPlanes` (1/3/5),
+`loadBalancer`, `autoscale.{enabled,minWorkers,maxWorkers,cpuScaleUpPercent,
+memScaleUpPercent,cpuScaleDownPercent,memScaleDownPercent,samplesToScale,
+cooldownMinutes}`. Missing keys are left alone.
+
+- `helpers/plan.ts` `planClusterChanges(current, desired)` is pure (unit
+  tested): create in the desired shape (workers, control planes, load
+  balancer at provision time) then autoscale; otherwise topology → workers
+  → control planes → load balancer → autoscale (only the changed keys).
+- `workers` is skipped while `autoscale.enabled` (the autoscaler owns it);
+  converting to multi-node already creates the workers.
+- Destructive steps (to single-node, removing the load balancer) need
+  `--yes`; without it they are printed and skipped.
+- `validateDesired` mirrors the server's bounds; a cluster that is not
+  `ready` is refused (apply again once it settles).
+
+## Files
+
+| Path | Responsibility |
+|---|---|
+| `src/commands/cluster/index.ts` | registration (`--cluster` / `--clusterName` on every one) |
+| `src/commands/cluster/{status,health,scale-events,cost,scale,scale-control-planes,autoscale,multi-node,single-node,load-balancer}.ts` | actions |
+| `src/commands/cluster/helpers/cluster-change.ts` | target resolution, int validation, mutate → print → `--wait` (`pollUntilReady`; ends in `error` → non-zero exit) |
+| `src/commands/cluster/types.ts` | local types + selections (newer than `@introspection`) |
+| `src/services/gql-client.ts` | the API calls |
+
+## How to test
+
+`npx jest --coverage=false test/cluster-scaling.spec.ts`; manually against a
+project with a private cluster: `gcli cluster:status`, then
+`gcli cluster:scale --clusterName <n> --workers 2 --wait`.
+
+## How to debug
+
+| Symptom | Look at |
+|---|---|
+| "Pass --cluster <id> or --clusterName" | no cluster given; `gcli cluster:list` |
+| autoscale refused | the server's message: single-node clusters must `cluster:multi-node` first; external clusters cannot autoscale |
+| `--wait` ends in error | `gcli cluster:list` (status + error), the dashboard's cluster page |
+
+## Not in the CLI yet
+
+Resource migration between clusters (`migrateResourcesToCluster` + status /
+retry), Fission options, the cluster agent install/rotate - dashboard only.

package/docs/features/lambforge-agent-skill.md

@@ -0,0 +1,22 @@
+# lambforge-agent-skill (gcli slice)
+
+**Canonical doc:** `gql-fission/docs/features/lambforge-agent-skill.md` ·
+**Since:** 2026-10-19
+
+The Lambforge agent skill (entry https://lambforge.com/skills/platform.md,
+topic files next to it; source in gql-fission `src/skills/`) teaches AI
+agents to drive this CLI. `cli.md` lists every command, and `platform.md`,
+`deploy.md`, `events.md`, `clusters.md` and `agents.md` quote gcli exactly:
+
+- install: `npm i -g @gapi/gcli` (or `npx @gapi/gcli`);
+- login: the command from Profile -> CLI (`login --token --key --url
+  --uploadUrl`), run by the human;
+- `project:list` / `project:use`;
+- `mcp:start` without `--url` (proxy to `<login url>/mcp`, token refresh on
+  401/403) as the MCP server: `claude mcp add --scope user lambforge -- gcli mcp:start`;
+- the `lambda:*`, `environment:*`, `cron:*`, `cluster:*`, `rabbitmq:*`,
+  `plugin:*`, `build`, `start` commands and flags it lists.
+
+**When you change any of those commands, flags or `mcp:start`, update
+the skill files in gql-fission in the same change** - agents copy it verbatim.
+README.md "AI agent skill" points users to it.

package/docs/features/local-ai-runners.md

@@ -13,6 +13,8 @@ what lives in gcli.
 | `gcli runner:list` | Your runners: online, last seen, models |
 | `gcli runner:models [--id] [--show a,b \| --all \| --none]` | Which models appear in the chat picker; no flag = list |
 | `gcli runner:delete [--id]` | Delete a runner and revoke its token (default: the saved one) |
+| `gcli runner:daemon:install` (alias `runner:start:daemon`) `[runner:start options] [--service] [--dry-run]` | Resolve/register the runner now (saves `runner.json`), then install + enable + restart a systemd USER unit running `runner:start --ollama <resolved>` (Linux) |
+| `gcli runner:daemon:status` / `:logs` / `:uninstall` | `systemctl --user status`, `journalctl --user -f`, disable --now + remove the unit |
 
 Endpoints: `--wsUrl` > `runner.json` > derived from the `gcli login` URL
 (`https://api.lambforge.com` -> `wss://api.lambforge.com/subscriptions`;
@@ -42,6 +44,15 @@ Ollama: `--ollama` > `$OLLAMA_HOST` (scheme optional, `0.0.0.0` -> `127.0.0.1`)
   The backend fits every request to it, because Ollama silently drops the
   oldest messages of a prompt that is too long (see the canonical doc,
   "Context window").
+- `daemon.ts` - `buildUnit` (pure; systemd quoting; `Restart=always`,
+  `RestartPreventExitStatus=78`), `resolveGcliCommand` (node + script, or
+  the SEA binary; warns for the npx cache), `SystemdUserService`
+  (`systemctl --user` via execFile, never a shell; home dir injectable so
+  tests can't touch the real `~/.config`). `daemon-commands.ts` - the
+  `runner:daemon:*` actions. Adapted from cli-builder's `SystemctlService`,
+  but a USER unit (no root) and no secrets in the unit. `runner:start` exits
+  78 (EX_CONFIG) when the token is refused, so systemd stops instead of
+  restart-looping.
 - `config.ts` (saved runner, URL derivation), `start.ts`, `list.ts`,
   `models.ts`, `delete.ts`, `index.ts` (command registration), `types.ts`.
 - `src/services/gql-client.ts` - `createLocalRunner`, `listLocalRunners`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gapi/gcli",
-  "version": "1.8.239",
+  "version": "1.8.241",
   "repository": {
     "type": "git",
     "url": "https://github.com/Stradivario/gapi.git"