AnythingLLM Custom Agent Skill Builder

When to Use

Use this skill when the user asks to create, build, scaffold, or generate an AnythingLLM custom agent skill. This includes requests like:

• "Create an AnythingLLM skill that..."

• "Build an agent skill for AnythingLLM"

• "Add a custom skill to my AnythingLLM"

• "Make my AnythingLLM agent able to..."

User's AnythingLLM Setup

• EasyPanel service: basheer / anything-llm

• Base API URL: https://basheer-anything-llm.c9tnyg.easypanel.host/api/v1

• Docker volume: storage mounted at /app/server/storage

• Skills path (inside container): /app/server/storage/plugins/agent-skills/

• API key env: ANYTHING_LLM_KEY (in ANC app's .env )

• Hot reload: Skills are hot-loaded — no restart needed. Just exit active agent session and reload page.

• AnythingLLM version requirement: Docker v1.2.2+ (user has this)

Skill File Structure

Every AnythingLLM custom skill is a folder containing exactly these files:

plugins/agent-skills/{hubId}/ ├── plugin.json # Metadata, params, setup args ├── handler.js # Runtime logic (Node.js) └── README.md # Optional documentation

CRITICAL: The folder name MUST match the hubId in plugin.json exactly.

plugin.json Schema (skill-1.0.0)

{ "active": true, "name": "Human Readable Skill Name", "hubId": "kebab-case-folder-name", "schema": "skill-1.0.0", "version": "1.0.0", "description": "What this skill does — the LLM reads this to decide when to invoke it", "author": "@basheer", "author_url": "https://github.com/khaledbashir", "license": "MIT", "setup_args": { "API_KEY": { "type": "string", "required": false, "input": { "type": "text", "default": "", "placeholder": "sk-xxxxx", "hint": "Optional API key for the service" }, "value": "" } }, "examples": [ { "prompt": "Example user prompt that should trigger this skill", "call": "{"param1": "value1", "param2": "value2"}" } ], "entrypoint": { "file": "handler.js", "params": { "param1": { "description": "What this parameter is for", "type": "string" } } }, "imported": true }

Field Reference

Field Type Required Notes

active

boolean YES Set true to load the skill

name

string YES Human-readable name shown in UI

hubId

string YES Must match folder name exactly (kebab-case)

schema

string YES Always "skill-1.0.0"

version

string YES Semver (e.g., "1.0.0" )

description

string YES LLM reads this to decide invocation — be specific

author

string no "@basheer"

author_url

string no "https://github.com/khaledbashir"

license

string no "MIT"

setup_args

object no UI-configurable settings (API keys, URLs, etc.)

examples

array no Few-shot examples help the LLM know when to call this

entrypoint

object YES file

• params the handler expects

imported

boolean YES Always true for custom skills

entrypoint.params types

• "string" — text input

• "number" — numeric input

• "boolean" — true/false

setup_args input types

• "text" — standard text field

• Each setup_arg needs: type , required , input.type , input.default , input.placeholder , input.hint

handler.js Template

// {skill-name} — AnythingLLM Custom Agent Skill // Created for ANC Proposal Engine

module.exports.runtime = { handler: async function ({ param1, param2 }) { try { this.introspect(Processing: ${param1}...);

  // Access setup_args (configured in UI):
  // const apiKey = this.runtimeArgs["API_KEY"];

  // Access skill metadata:
  // const { name, hubId, version } = this.config;

  // Your logic here...
  const result = await fetch("https://api.example.com/endpoint", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({ param1, param2 }),
  });

  const data = await result.json();

  // MUST return a string — anything else breaks the agent
  return JSON.stringify(data, null, 2);
} catch (e) {
  this.introspect(`Error: ${e.message}`);
  this.logger("Skill error", e.message);
  return `Skill failed: ${e.message}`;
}

}, };

handler.js Rules

• Must export module.exports.runtime with a handler async function

• Must return a string — anything else breaks the agent or loops forever

• Must wrap everything in try/catch — return error as string

• this.introspect(msg) — shows thinking text in the AnythingLLM UI

• this.logger(label, msg) — writes to server console for debugging

• this.runtimeArgs — object of setup_args values from plugin.json

• this.config — { name, hubId, version } metadata

• Parameters come from entrypoint.params — destructure from the single argument object

• require() modules inside the handler — not at top level

• Bundle dependencies — node_modules must be inside the skill folder if needed

Deployment Instructions

After generating the skill files, tell the user:

Copy the skill folder to AnythingLLM container:

From local machine or server

docker cp ./my-skill-folder basheer-anything-llm:/app/server/storage/plugins/agent-skills/

Or via EasyPanel terminal:

Inside the container, skills live at:

/app/server/storage/plugins/agent-skills/

Reload AnythingLLM — exit any active agent chat, refresh the browser

Enable the skill — go to Agent Settings > Skills, find the new skill, toggle it on

Configure setup_args — if the skill has setup_args, click the gear icon to set values

Test — start a new agent chat and try one of the example prompts

Output Location

Generate skill files at: /root/rag2/.claude/generated-skills/{hubId}/

• This staging area lets the user review before deploying

• Include a deploy.sh script for easy container copy

Common Skill Patterns

API Proxy Skill

Wraps an external REST API so the AnythingLLM agent can call it conversationally.

Database Query Skill

Queries a database and returns formatted results. For ANC, could query the proposal DB via the Next.js API.

ANC-Specific Skills

When building skills that interact with the ANC Proposal Engine:

• API base: Use setup_arg for the URL (default: https://basheer-natalia.prd42b.easypanel.host )

• Auth: Include auth token in setup_args if the endpoint requires it

• Endpoints: See the ANC API routes in the anc-bible skill for available endpoints

Example: ANC Product Lookup Skill

// plugin.json { "active": true, "name": "ANC Product Catalog Search", "hubId": "anc-product-lookup", "schema": "skill-1.0.0", "version": "1.0.0", "description": "Search the ANC LED product catalog by manufacturer, pixel pitch, or environment. Returns matching products with specs and pricing.", "author": "@basheer", "license": "MIT", "setup_args": { "ANC_API_URL": { "type": "string", "required": true, "input": { "type": "text", "default": "https://basheer-natalia.prd42b.easypanel.host", "placeholder": "https://your-anc-instance.com", "hint": "Base URL of your ANC Proposal Engine" }, "value": "https://basheer-natalia.prd42b.easypanel.host" } }, "examples": [ { "prompt": "Find me LG outdoor LED panels with 4mm pitch", "call": "{"search": "LG", "environment": "outdoor", "pitchMax": "5"}" }, { "prompt": "What Yaham products do we have?", "call": "{"search": "Yaham"}" } ], "entrypoint": { "file": "handler.js", "params": { "search": { "description": "Text search across product name, model number, manufacturer", "type": "string" }, "environment": { "description": "Filter by environment: indoor, outdoor, or indoor_outdoor", "type": "string" }, "pitchMin": { "description": "Minimum pixel pitch in mm", "type": "string" }, "pitchMax": { "description": "Maximum pixel pitch in mm", "type": "string" } } }, "imported": true }

// handler.js module.exports.runtime = { handler: async function ({ search, environment, pitchMin, pitchMax }) { try { const baseUrl = this.runtimeArgs["ANC_API_URL"] || "https://basheer-natalia.prd42b.easypanel.host"; const params = new URLSearchParams(); if (search) params.set("search", search); if (environment) params.set("environment", environment); if (pitchMin) params.set("pitchMin", pitchMin); if (pitchMax) params.set("pitchMax", pitchMax);

  this.introspect(`Searching ANC product catalog: ${params.toString() || "all products"}...`);

  const response = await fetch(`${baseUrl}/api/products?${params.toString()}`);
  if (!response.ok) return `API error: ${response.status} ${response.statusText}`;

  const data = await response.json();
  if (!data.products || data.products.length === 0) {
    return "No products found matching your criteria.";
  }

  const summary = data.products.map(p =>
    `${p.manufacturer} ${p.displayName || p.modelNumber} — ${p.pixelPitch}mm pitch, ${p.environment}, ${p.maxNits || "N/A"} nits, ${p.cabinetWidthMm}x${p.cabinetHeightMm}mm cab`
  ).join("\n");

  return `Found ${data.total} products:\n${summary}`;
} catch (e) {
  this.introspect(`Failed to search products: ${e.message}`);
  return `Product search failed: ${e.message}`;
}

}, };