@n8n/instance-ai 1.26.3 → 1.27.1

package/skills/debugging-executions/SKILL.md

@@ -86,6 +86,10 @@ first move whenever the user is debugging a **read** node that already failed a
 real execution: the node ran for real once already, and a mock-only check is
 what sends the user back for a second session.
 
+The action refuses a replay that would re-run a node above the target: an
+unpinned one the execution never reached, one that failed there, or a loop
+that did not finish. Pick another execution, or tell the user which node would run again.
+
 ### Decide whether the node is safe to run first
 
 A step run is a real run. The node uses the user's real credentials and reaches
@@ -142,6 +146,55 @@ Mocked input does **not** make a write node safe. The node still runs for real
 against the user's systems; only its input is invented, which makes the effect
 less predictable, not more.
 
+### Running a tool
+
+A tool never runs on its own. n8n runs it through the node that owns it —
+usually the Agent. A step run on a tool therefore behaves like a step run on
+that Agent:
+
+- `mockInput` feeds the **Agent's** input, not the tool's arguments.
+- `reuseExecutionId` replays the nodes above the **Agent**.
+- A chain run (neither option) runs every node above the Agent for real. Give
+  one of the two options when a node up there writes.
+- `ranThroughNodeNames` lists **every** node that can run the tool, not the one
+  that ran it. A tool on several agents lists them all: the engine picks one,
+  and it does not report which. Do not name a single Agent to the user when
+  this field holds more than one.
+- A tool on two Agents where one runs above the other is refused. Run the
+  upper Agent, then read the tool with `executions(action="get-node-output")`.
+
+The tool's own arguments come from `toolArguments` — the values the agent would
+normally decide:
+
+```
+executions(action="run-step", workflowId, nodeName="Search Tickets Tool",
+           reuseExecutionId=<the failed execution>,
+           toolArguments={"query": "login fails", "status": "open"})
+```
+
+The example targets a **read** tool on purpose. The write rule above holds here
+too, and a tool hides the write behind a friendly name: a step run on a "Create
+Ticket" tool creates the ticket again, and `toolArguments` does not change that.
+Only run one when the user has accepted a second write.
+
+Use the argument names from the node's `$fromAI` calls, which
+`workflows(action="get-as-code")` shows. Pass a plain string instead for a tool
+that takes one free-text input (Wikipedia, Code Tool, a vector store used as a
+tool).
+
+A tool that declares `$fromAI` arguments is refused without them: it would
+otherwise fail for a reason that has nothing to do with the user's problem, and
+you would report that as the defect. A node that holds several tools is refused
+outright, because nothing here can name one of its tools the way the agent
+does: the "MCP Client Tool" node, and every node the MCP registry added, whose
+type is `@n8n/mcp-registry.<server slug>`. Run the Agent for those, and read
+the node's output from that execution.
+
+A sub-node that is not a tool — a model, memory, embeddings — cannot be run this
+way at all. Run the Agent, and read the sub-node with
+`executions(action="get-node-output")` on **that** execution: n8n records every
+call a sub-node made while the Agent ran.
+
 ## Successful execution with wrong or empty value
 
 When `debug` doesn't apply because nothing errored, call

package/skills/intent-recognition/SKILL.md

@@ -281,7 +281,7 @@ continuity for that primitive.
 
 **Unsupported capabilities**: when the user names a specific channel or
 capability for an agent (e.g. "WhatsApp", "Teams"), call
-`list-agent-capabilities` before classifying. If the named channel is
+`agent-context` with `type: "capabilities"` before classifying. If the named channel is
 absent, it is unsupported for agents — do not classify the request as a
 workflow substitute, do not improvise workflow nodes to fake the channel,
 and do not claim it can be configured. Explain that it is unavailable for
@@ -396,7 +396,7 @@ valid, apply the growth tiebreaker instead of asking a theoretical preference.
   actions or whether an investigator must choose and adapt them.
 - "Build me an agent my team can @mention on WhatsApp to triage customer
   messages." -> **agent-anchored** (explicit agent artifact + chat
-  interaction), but call `list-agent-capabilities` first: WhatsApp is absent,
+  interaction), but call `agent-context` with `type: "capabilities"` first: WhatsApp is absent,
   so do not build. Explain WhatsApp is unsupported for agents, offer the
   supported chat channels the tool returned, with their
   `capabilities`, and ask which to use — or whether the user wants a
@@ -438,7 +438,7 @@ valid, apply the growth tiebreaker instead of asking a theoretical preference.
   workflow-anchored pipeline. A Chat Trigger workflow is correct only when
   chat is merely the manual trigger for a fixed graph.
 - Never improvise a workflow substitute for an unsupported agent channel or
-  capability. When the user names a channel not in `list-agent-capabilities`,
+  capability. When the user names a channel not returned by `agent-context`,
   explain the limitation and offer supported alternatives — do not add
   workflow nodes that fake the channel or silently translate the request
   into a workflow change.

package/skills/workflow-builder/SKILL.md

@@ -59,6 +59,20 @@ first action turn (each extra sequential turn resends the whole context). When
 unsure which nodes to use, load this skill first and follow its research
 process below.
 
+## Early service connections
+
+When the `credentials(action="setup")` schema offers `filePath`, announce
+known service credentials immediately after the first node definitions or
+credential type search. Do this before detailed planning, SDK research, or
+source generation. Do not wait for every service to be resolved.
+
+Pick the source `filePath` and call credential setup with the complete list
+of types known so far. Make it the only tool call in that response. Wait for
+its successful `preBuild: true` result before continuing. Add later discoveries
+with another complete list. Do not batch setup with source writing.
+
+When early setup is unavailable, keep the post-build setup flow below.
+
 ## Repair Strategy
 
 When the edit is to fix a node the user reports as erroring or showing a red
@@ -97,10 +111,12 @@ resources, credentials, channel IDs, or timezone; use placeholders or unresolved
 for a capability the user did not name,
 discover coverage first and use a Gateway credits–covered node instead of asking
 when the user has no credential for a comparable tool (see Gateway credits
-Preference). Setup details — recipients, accounts,
-resources, channels, credentials, timezone — belong in placeholders or
-unresolved `newCredential()` calls until post-build setup. After the first
-build, use `ask-user` when stuck or genuinely ambiguous; do not retry the same
+Preference). Setup details such as recipients, accounts, resources,
+channels, credentials, and timezone belong in placeholders or
+unresolved `newCredential()` calls. Announce known credentials through early
+setup when available, then continue building without waiting for the user.
+After the first build, resolve remaining setup and use `ask-user` when stuck
+or when choices are ambiguous; do not retry the same
 failing approach more than twice. Never re-ask an answered, deferred, or skipped
 question. A skip grants no additional permission. Choose defaults only for
 unspecified details within the requested task. If a skipped question seeks
@@ -978,7 +994,7 @@ For AI Agent workflows:
 
 After building a workflow that uses a trigger with an HTTP endpoint, share the
 full production URL with the user. Use the Webhook base URL and Form base URL
-from Instance Info in the system prompt. Each trigger type has a distinct
+from the `<instance-urls>` block in the user's turn. Each trigger type has a distinct
 pattern:
 
 - **Webhook Trigger**: `{webhookBaseUrl}/{path}` (where `{path}` is the node's

package/dist/tools/agents.tool.d.ts

@@ -1,2 +0,0 @@
-import type { OrchestrationContext } from '../types';
-export declare function createAgentsTool(context: OrchestrationContext): import("@n8n/agents").BuiltTool;

package/dist/tools/agents.tool.js

@@ -1,40 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.createAgentsTool = createAgentsTool;
-const agents_1 = require("@n8n/agents");
-const zod_1 = require("zod");
-const tool_ids_1 = require("./tool-ids");
-const agentsInputSchema = zod_1.z.object({
-    action: zod_1.z
-        .literal('list')
-        .describe('List the n8n Agents in this project (id, name, published state, last update), most recently updated first.'),
-});
-const agentsOutputSchema = zod_1.z.object({
-    count: zod_1.z.number(),
-    agents: zod_1.z.array(zod_1.z.object({
-        agentId: zod_1.z.string(),
-        name: zod_1.z.string(),
-        published: zod_1.z.boolean(),
-        updatedAt: zod_1.z.string(),
-    })),
-});
-function createAgentsTool(context) {
-    return new agents_1.Tool(tool_ids_1.DOMAIN_TOOL_IDS.AGENTS)
-        .description('List the n8n Agent artifacts in this project — id, name, published state, and last ' +
-        'update, most recently updated first. Use it to answer questions about existing ' +
-        'agents and to find the `agentId` to pass to `build-agent` when the user wants to ' +
-        'edit an agent that was not built in this conversation. Read-only; agents are ' +
-        'created and edited via `build-agent`.')
-        .input(agentsInputSchema)
-        .output(agentsOutputSchema)
-        .handler(async () => {
-        const delegate = context.domainContext?.builderDelegate;
-        if (!delegate) {
-            throw new Error('Agent listing is not available on this instance.');
-        }
-        const agents = await delegate.listAgents();
-        return { count: agents.length, agents };
-    })
-        .build();
-}
-//# sourceMappingURL=agents.tool.js.map

package/dist/tools/agents.tool.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"agents.tool.js","sourceRoot":"","sources":["../../src/tools/agents.tool.ts"],"names":[],"mappings":";;;AAMA,wCAAmC;AACnC,6BAAwB;AAGxB,yCAA6C;AAE7C,MAAM,iBAAiB,GAAG,OAAC,CAAC,MAAM,CAAC;IAClC,MAAM,EAAE,OAAC;SACP,OAAO,CAAC,MAAM,CAAC;SACf,QAAQ,CACR,4GAA4G,CAC5G;CACF,CAAC,CAAC;AAEH,MAAM,kBAAkB,GAAG,OAAC,CAAC,MAAM,CAAC;IACnC,KAAK,EAAE,OAAC,CAAC,MAAM,EAAE;IACjB,MAAM,EAAE,OAAC,CAAC,KAAK,CACd,OAAC,CAAC,MAAM,CAAC;QACR,OAAO,EAAE,OAAC,CAAC,MAAM,EAAE;QACnB,IAAI,EAAE,OAAC,CAAC,MAAM,EAAE;QAChB,SAAS,EAAE,OAAC,CAAC,OAAO,EAAE;QACtB,SAAS,EAAE,OAAC,CAAC,MAAM,EAAE;KACrB,CAAC,CACF;CACD,CAAC,CAAC;AAEH,0BAAiC,OAA6B;IAC7D,OAAO,IAAI,aAAI,CAAC,0BAAe,CAAC,MAAM,CAAC;SACrC,WAAW,CACX,qFAAqF;QACpF,iFAAiF;QACjF,mFAAmF;QACnF,+EAA+E;QAC/E,uCAAuC,CACxC;SACA,KAAK,CAAC,iBAAiB,CAAC;SACxB,MAAM,CAAC,kBAAkB,CAAC;SAC1B,OAAO,CAAC,KAAK,IAAI,EAAE;QACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,aAAa,EAAE,eAAe,CAAC;QACxD,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;QACrE,CAAC;QACD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,UAAU,EAAE,CAAC;QAC3C,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;IACzC,CAAC,CAAC;SACD,KAAK,EAAE,CAAC;AACX,CAAC"}

package/dist/tools/orchestration/list-agent-capabilities.tool.d.ts

@@ -1,2 +0,0 @@
-import type { OrchestrationContext } from '../../types';
-export declare function createListAgentCapabilitiesTool(context: OrchestrationContext): import("@n8n/agents").BuiltTool;

package/dist/tools/orchestration/list-agent-capabilities.tool.js

@@ -1,45 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.createListAgentCapabilitiesTool = createListAgentCapabilitiesTool;
-const agents_1 = require("@n8n/agents");
-const zod_1 = require("zod");
-const tool_ids_1 = require("../tool-ids");
-const chatIntegrationDescriptorSchema = zod_1.z.object({
-    type: zod_1.z.string(),
-    label: zod_1.z.string(),
-    icon: zod_1.z.string(),
-    credentialTypes: zod_1.z.array(zod_1.z.string()),
-    capabilities: zod_1.z.array(zod_1.z.string()).optional(),
-    useIntegrationWhen: zod_1.z.array(zod_1.z.string()).optional(),
-    useNodeToolWhen: zod_1.z.array(zod_1.z.string()).optional(),
-});
-const listAgentCapabilitiesOutputSchema = zod_1.z.object({
-    channels: zod_1.z.array(chatIntegrationDescriptorSchema),
-    agentCapabilities: zod_1.z.array(zod_1.z.string()),
-    limitations: zod_1.z.array(zod_1.z.string()),
-});
-function createListAgentCapabilitiesTool(context) {
-    return new agents_1.Tool(tool_ids_1.ORCHESTRATION_TOOL_IDS.LIST_AGENT_CAPABILITIES)
-        .description('List what n8n Agents can do and the chat-channel integrations they support. ' +
-        "Returns each supported channel's `type`, `label`, supported credential types, " +
-        'and builder guidance (`capabilities`, `useIntegrationWhen`, `useNodeToolWhen`), ' +
-        'plus `agentCapabilities` (what an agent can do beyond chat) and `limitations` ' +
-        '(agent-level constraints to respect when planning a build). ' +
-        'Call this before building or modifying an agent whenever the user names a ' +
-        'specific channel or capability (e.g. WhatsApp, Teams): if the named channel ' +
-        'is absent from `channels`, it is unsupported for agents — explain the ' +
-        'limitation and offer the listed alternatives instead of improvising a ' +
-        'workflow substitute or claiming it is configured. Read-only; channels and ' +
-        'capabilities are configured via `build-agent`.')
-        .input(zod_1.z.object({}))
-        .output(listAgentCapabilitiesOutputSchema)
-        .handler(async () => {
-        const delegate = context.domainContext?.builderDelegate;
-        if (!delegate) {
-            throw new Error('Agent capabilities are not available on this instance.');
-        }
-        return await delegate.listAgentCapabilities();
-    })
-        .build();
-}
-//# sourceMappingURL=list-agent-capabilities.tool.js.map

package/dist/tools/orchestration/list-agent-capabilities.tool.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"list-agent-capabilities.tool.js","sourceRoot":"","sources":["../../../src/tools/orchestration/list-agent-capabilities.tool.ts"],"names":[],"mappings":";;;AAWA,wCAAmC;AACnC,6BAAwB;AAGxB,0CAAqD;AAErD,MAAM,+BAA+B,GAAG,OAAC,CAAC,MAAM,CAAC;IAChD,IAAI,EAAE,OAAC,CAAC,MAAM,EAAE;IAChB,KAAK,EAAE,OAAC,CAAC,MAAM,EAAE;IACjB,IAAI,EAAE,OAAC,CAAC,MAAM,EAAE;IAChB,eAAe,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC;IACpC,YAAY,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC5C,kBAAkB,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAClD,eAAe,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC/C,CAAC,CAAC;AAEH,MAAM,iCAAiC,GAAG,OAAC,CAAC,MAAM,CAAC;IAElD,QAAQ,EAAE,OAAC,CAAC,KAAK,CAAC,+BAA+B,CAAC;IAElD,iBAAiB,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC;IAEtC,WAAW,EAAE,OAAC,CAAC,KAAK,CAAC,OAAC,CAAC,MAAM,EAAE,CAAC;CAChC,CAAC,CAAC;AAEH,yCAAgD,OAA6B;IAC5E,OAAO,IAAI,aAAI,CAAC,iCAAsB,CAAC,uBAAuB,CAAC;SAC7D,WAAW,CACX,8EAA8E;QAC7E,gFAAgF;QAChF,kFAAkF;QAClF,gFAAgF;QAChF,8DAA8D;QAC9D,4EAA4E;QAC5E,8EAA8E;QAC9E,wEAAwE;QACxE,wEAAwE;QACxE,4EAA4E;QAC5E,gDAAgD,CACjD;SACA,KAAK,CAAC,OAAC,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;SACnB,MAAM,CAAC,iCAAiC,CAAC;SACzC,OAAO,CAAC,KAAK,IAAI,EAAE;QACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,aAAa,EAAE,eAAe,CAAC;QACxD,IAAI,CAAC,QAAQ,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;QAC3E,CAAC;QAID,OAAO,MAAM,QAAQ,CAAC,qBAAqB,EAAE,CAAC;IAC/C,CAAC,CAAC;SACD,KAAK,EAAE,CAAC;AACX,CAAC"}