@desplega.ai/agent-swarm 2.7.4 → 2.7.6

package/templates/skills/script-workflows/content.md

@@ -43,7 +43,25 @@ export default async function main(args, ctx) {
 }
 ```
 
-`ctx.step.agentTask` blocks until the dispatched task reaches a terminal status (default `waitForCompletion: true`, default `timeoutMs` 2h) and journals its real output — a sequential `plan → implement → review` chain will not fan out. Pass `waitForCompletion: false` for the legacy fire-and-poll-yourself shape, or `failOnTaskFailure: false` to receive `{taskId, status: "failed", error}` instead of a throw when the task fails/is cancelled.
+`ctx.step.agentTask` blocks until the dispatched task reaches a terminal status (default `waitForCompletion: true`, default `timeoutMs` 2h) and journals its real output — a sequential `plan → implement → review` chain will not fan out. Pass `failOnTaskFailure: false` to receive `{taskId, status: "failed", error}` instead of a throw when the task fails/is cancelled.
+
+A run launched from a `follow-up` task in a Slack thread passes the same re-delegation guard as `send-task`, decided once for the whole run at its first new step: a step that repeats completed thread work, or a run past the thread's dispatch cap (a run counts once), throws with the rule's message, and every later new step of that run throws too. Pass `allowRedelegation: true` on a step only for a deliberate re-run. Do not wrap a single dispatch in a durable run; use `send-task`.
+
+`waitForCompletion: false` does not return right after dispatch. The first call still waits up to ~30s while the server polls the task, then returns one of three shapes:
+
+- The task finished in that window: its output string or `null`, as on success.
+- The task failed or was cancelled in that window: a throw, or `{taskId, status: "failed", error}` with `failOnTaskFailure: false`.
+- The task is still running: `{taskId, status}`.
+
+On success, `ctx.step.agentTask` resolves to the worker's output string (JSON text under `outputSchema`, so parse it yourself; `null` when the task stored none), not a task record. On a string, `step.taskId` and `step.status` are undefined, so a guard such as `step.status !== "completed"` or `step?.taskId ?? "unknown"` silently misfires. On `null`, `step.taskId` throws a TypeError. Only `failOnTaskFailure: false` with a failed or cancelled task returns `{taskId, status: "failed", error}`, and only `waitForCompletion: false` on a task still running returns `{taskId, status}`. To get a step's task id, query by its context key (the label is the step label you passed):
+
+```sql
+SELECT id, status, output FROM agent_tasks
+WHERE contextKey = 'script-run:<runId>:<label>' AND taskType = 'script-run-step'
+ORDER BY createdAt DESC, rowid DESC LIMIT 1
+```
+
+Inside the run, `<runId>` is `ctx.run.id`.
 
 ## Label Rules
 

package/templates/skills/swarm-docs/content.md

@@ -91,12 +91,14 @@ Fetch the page as markdown: `curl -fsSL https://docs.agent-swarm.dev/llms.mdx/do
 - [One-off Script Workflow Runs](https://docs.agent-swarm.dev/docs/guides/script-workflow-runs). Reference.
 - [Dreaming](https://docs.agent-swarm.dev/docs/guides/dreaming). When: Someone asks what the nightly dreaming run changed, how to configure or turn it off, or how to undo one of its changes.
 - [Claude Bridge](https://docs.agent-swarm.dev/docs/guides/claude-bridge-experimental). Reference.
+- [Reading claude.ai Artifacts headless](https://docs.agent-swarm.dev/docs/guides/claude-artifacts-read). When: Someone wants an agent to read a claude.ai Artifact link, such as a design and its canvas.json, from a headless session.
 - [Worker Credential Recovery](https://docs.agent-swarm.dev/docs/guides/worker-credential-recovery). Reference.
 - [Model Gateways](https://docs.agent-swarm.dev/docs/guides/provider-auth/model-gateways). Reference.
 - [Codex OAuth](https://docs.agent-swarm.dev/docs/guides/provider-auth/codex-oauth). Reference.
 - [Cursor](https://docs.agent-swarm.dev/docs/guides/provider-auth/cursor). Reference.
 - [Amp](https://docs.agent-swarm.dev/docs/guides/provider-auth/amp). Reference.
 - [Grok](https://docs.agent-swarm.dev/docs/guides/provider-auth/grok). Reference.
+- [fx](https://docs.agent-swarm.dev/docs/guides/provider-auth/fx). Reference.
 - [DeepSeek Harness (dsh)](https://docs.agent-swarm.dev/docs/guides/provider-auth/dsh). Reference.
 - [x402 Payments](https://docs.agent-swarm.dev/docs/guides/x402-payments). Reference.
 - [Multi-runtime agents](https://docs.agent-swarm.dev/docs/guides/multi-runtime-agents). Reference.

package/templates/skills/swarm-scripts/SKILL.md

@@ -23,6 +23,8 @@ Named script only when the logic will run two or more times, by you, another age
 
 Reference point: a workflow triage that took about 26 tool calls returns as one result of about 4k tokens in about 13 seconds.
 
+Pick the run mode before you author. A short inline `script-run` is for small lookups only. A job that scans a whole population or may run past the inline cap of about 30 s starts as a durable `launch-script-run` from the start. A single `ctx.step.swarmScript` step has the same cap, so collect in bounded steps: each step takes a cursor, collects one slice in well under 30 s, and returns its rows plus the next cursor. Label each step with its slice (`collect-${page}`) so the journal checkpoints it, then compute from the collected slices in a final step. Restarting a long inline run throws away the work it already did.
+
 ## Loading the tools
 
 The script tools are deferred. Load them with your harness tool search before the first call: `script-search`, `script-run`, `script-upsert`, `script-query-types`, `script-delete`, and for durable runs `launch-script-run`, `get-script-run`, `list-script-runs`.

package/templates/skills/swarm-scripts/content.md

@@ -18,6 +18,8 @@ Named script only when the logic will run two or more times, by you, another age
 
 Reference point: a workflow triage that took about 26 tool calls returns as one result of about 4k tokens in about 13 seconds.
 
+Pick the run mode before you author. A short inline `script-run` is for small lookups only. A job that scans a whole population or may run past the inline cap of about 30 s starts as a durable `launch-script-run` from the start. A single `ctx.step.swarmScript` step has the same cap, so collect in bounded steps: each step takes a cursor, collects one slice in well under 30 s, and returns its rows plus the next cursor. Label each step with its slice (`collect-${page}`) so the journal checkpoints it, then compute from the collected slices in a final step. Restarting a long inline run throws away the work it already did.
+
 ## Loading the tools
 
 The script tools are deferred. Load them with your harness tool search before the first call: `script-search`, `script-run`, `script-upsert`, `script-query-types`, `script-delete`, and for durable runs `launch-script-run`, `get-script-run`, `list-script-runs`.

package/templates/workflows/dreaming/content.md

@@ -1,6 +1,6 @@
 # Dreaming
 
-Nightly self-improvement loop. The `dreaming-daily` schedule triggers it once a day. A manual trigger with `{"dryRun": true}` (optionally `"agentIds": [...]`) runs it read-only and skips Slack.
+Nightly self-improvement loop. The `dreaming-daily` workflow schedule starts it once a day, as the lead. A 20-hour cooldown keeps it to one run a day: a trigger within 20 hours of the last run start (any status, manual and dry runs included) is recorded as a skipped run. It counts from starts only, so a long run never pushes out the next night, and two simultaneous triggers start one run. A manual trigger with `{"dryRun": true}` (optionally `"agentIds": [...]`) runs it read-only and skips Slack.
 
 How a night runs:
 
@@ -27,6 +27,10 @@ Trim backups and heartbeat drafts use the assigned worker's default agent-fs dri
 {
   "name": "dreaming",
   "description": "Nightly self-improvement loop. gather -> per-agent self-reports (foreach) -> Lead synthesis -> apply, trim, heartbeat, skills, scriptify and memory lanes -> verify -> learn -> publish -> optional Slack receipt. Writes that need Lead authority run as Lead agent-tasks. Manual trigger {dryRun:true, agentIds?:[...]} runs read-only and skips Slack.",
+  "cooldown": {
+    "hours": 20,
+    "from": "start"
+  },
   "input": {
     "v3Writes": "off",
     "leadAgentId": "",

package/dist/acp-adapter-96dc64m2.js

@@ -1,15 +0,0 @@
-import {
-  ACPAdapter2
-} from "./cli-avgcz4ra.js";
-import"./cli-13jvmew3.js";
-import"./cli-224epwwf.js";
-import"./cli-9cctx1tn.js";
-import"./cli-4bd1rck6.js";
-import"./cli-jc1g9ndz.js";
-import"./cli-j4kt1mpb.js";
-import"./cli-0726awxw.js";
-import"./cli-h55y2bjc.js";
-import"./cli-vqgnyh76.js";
-export {
-  ACPAdapter2 as ACPAdapter
-};

package/dist/cli-dznrgaep.js

@@ -1,30 +0,0 @@
-import {
-  require_permessage_deflate,
-  require_receiver,
-  require_sender,
-  require_extension,
-  require_websocket,
-  require_stream,
-  require_subprotocol,
-  require_websocket_server
-} from "./cli-hf6zt5m6.js";
-import {
-  __toESM,
-  __esm
-} from "./cli-242037k3.js";
-
-// node_modules/ws/wrapper.mjs
-var import_stream, import_extension, import_permessage_deflate, import_receiver, import_sender, import_subprotocol, import_websocket, import_websocket_server, wrapper_default;
-var init_wrapper2 = __esm(() => {
-  import_stream = __toESM(require_stream(), 1);
-  import_extension = __toESM(require_extension(), 1);
-  import_permessage_deflate = __toESM(require_permessage_deflate(), 1);
-  import_receiver = __toESM(require_receiver(), 1);
-  import_sender = __toESM(require_sender(), 1);
-  import_subprotocol = __toESM(require_subprotocol(), 1);
-  import_websocket = __toESM(require_websocket(), 1);
-  import_websocket_server = __toESM(require_websocket_server(), 1);
-  wrapper_default = import_websocket.default;
-});
-
-export { import_websocket, import_websocket_server, wrapper_default, init_wrapper2 };