developersai-skillsextensibilitytutorialagentic-ai

Building Custom AI Skills in aiFetchly: Developer Guide

September 30, 2026

aiFetchly System Settings Manage AI Skills table displaying installed extensions, permission tiers, and manifest schema viewer

Expanding Desktop AI with Modular Tooling

While frontier large language models exhibit powerful reasoning, their operational utility in production software workflows depends entirely on external tooling. Standard chat interfaces restrict developers to vendor-bundled tools or require maintaining brittle, cloud-hosted webhook endpoints.

AI Fetchly introduces a native desktop extension architecture called AI Skills. Skills are self-contained, versioned tool packages written in JavaScript or Python that give AI Chat deterministic capabilities—such as querying internal databases, transforming niche data formats, or scraping authenticated endpoints.

By defining an explicit manifest.json schema and permission boundaries, developers can deploy custom tools locally without running remote servers or leaking proprietary business logic.

Technical architecture diagram of the aiFetchly custom AI Skill lifecycle from manifest authoring to permission gating and in-chat execution

Permission Categories and Approval Gates

To maintain workstation security, aiFetchly categorizes every installed skill based on its declared permissions:

Permission Category	Operational Capabilities	Approval Behavior in AI Chat
`pure`	Text formatting, regex transformations, calculations	Always auto-approved without user prompts
`network`	Outbound HTTP requests, external API queries	Prompts by default; auto-approves in Full Access mode
`filesystem`	Local workspace read and write operations	Prompts with file path preview before execution
`automation`	Browser automation, DOM interaction, scraping	Prompts with target URL and script scope
`shell`	Arbitrary OS shell execution (built-in only)	Prompts on every single command; never permanently bypassed

Step-by-Step Tutorial: Creating and Deploying a Custom Skill

Follow this walkthrough to author, validate, and install a custom API data fetcher skill.

Step 1: Define the `manifest.json` Specification

In your project directory, create a manifest.json file. The manifest informs the AI model about what the skill accomplishes and provides the JSON Schema for input validation:

{
  "name": "custom-api-fetcher",
  "version": "1.0.0",
  "description": "Fetches and sanitizes JSON records from external analytics endpoints.",
  "runtime": "javascript",
  "entry": "index.js",
  "parameters": {
    "type": "object",
    "properties": {
      "endpointUrl": { "type": "string" },
      "timeoutMs": { "type": "number" }
    },
    "required": ["endpointUrl"]
  },
  "permissions": ["network"]
}

Note: Permissions strictly accept network, filesystem, or automation. If omitted, the skill defaults to pure.

Step 2: Implement the Entry Logic

Create index.js adjacent to your manifest:

module.exports = async function run(params) {
  const { endpointUrl, timeoutMs = 5000 } = params;
  const controller = new AbortController();
  const id = setTimeout(() => controller.abort(), timeoutMs);

  try {
    const response = await fetch(endpointUrl, { signal: controller.signal });
    clearTimeout(id);
    const data = await response.json();
    return { success: true, count: data.length, data };
  } catch (err) {
    return { success: false, error: err.message };
  }
};

Step 3: Package and Import into aiFetchly

Compress the contents of your folder (not the outer directory) into custom-api-fetcher.zip.
Open aiFetchly, navigate to System Setting > Manage Skills.
Click the Import button in the upper toolbar and select your .zip archive.
Alternatively, copy the uncompressed folder directly to ~/.aifetchly/skills/custom-api-fetcher/ for automatic hot-reloading during development.

Once installed, the skill appears immediately in AI Chat V2 and Plan Mode, where the AI autonomously invokes it whenever your prompts demand external analytics data.

Safe, Sovereign Agentic Extensibility

By coupling strict JSON Schema parameter validation with transparent permission prompts, aiFetchly lets developers safely bridge autonomous reasoning with local and remote developer tools.

To explore pre-built skills and plugin management, visit the AI Fetchly Docs.

Visual Generation Prompts

Hero Image Prompt: Modern dark-mode developer UI showing the aiFetchly Manage Skills table in System Settings, displaying installed extensions with permission category chips (‘network’, ‘pure’, ‘filesystem’), an active manifest.json preview drawer, and an ‘Import Skill’ modal, sleek 3D isometric perspective, electric cyan and emerald accents.
Body Image Prompt: High-resolution technical pipeline flowchart illustrating the four-stage AI Skill execution lifecycle: manifest.json authoring, zip ingestion with permission validation, multi-tier approval gate routing (pure vs network vs filesystem), and sandboxed in-chat execution, clean slate vector styling.