Forge App Builder
When building a Forge app, the agent MUST complete this workflow in order. Do not skip steps. Do not substitute manual instructions for running the scripts below.
Critical Rules
- Always use
forge createto scaffold apps — it registers the app and generates a valid app ID - Never manually scaffold — apps without valid app IDs cannot be deployed
- If
forge createfails, STOP — inform the user and provide the manual command - Never ask for API tokens in chat — direct users to run
forge loginin their terminal and enter credentials there - Always ask the user to choose when multiple options exist (developer spaces, sites) — never pick on their behalf
- Always ask the user for their Atlassian site URL during installation — never try to discover it from other apps, environment variables, or any other source
- Always run the deploy script for deploy and install — do not give the user only manual
forge deploy/forge installcommands as the primary outcome; runscripts.deploy_forge_app.pyyourself
MCP Server Prerequisites
This skill works best with the following MCP servers. The scripts and CLI workflow function without them, but the agent will lack access to up-to-date Forge documentation for template selection, module configuration, and code patterns.
| MCP Server | Required? | Purpose |
|---|---|---|
| Forge MCP | Strongly recommended | Template lookup, module discovery, manifest syntax, UI Kit/backend guides |
| ADS MCP | Optional | Atlaskit component/token/icon lookup (Custom UI apps only) |
If MCP servers are not connected, inform the user that code guidance may be based on the model's training data and could be outdated. Recommend they verify against developer.atlassian.com/platform/forge. |
Agent Workflow
Complete steps 0–5 in order. Run the scripts yourself; do not only instruct the user to run them.
Step 0: Prerequisites (Install Automatically If Missing)
Before any other steps, call the forge-development-guide tool to get the current Node.js version requirement and CLI setup instructions. Then check and install prerequisites:
- Node.js — Run
node -v. If missing or below the version specified in the development guide:
- macOS (Homebrew):
brew install node - nvm:
nvm install <version>thennvm use <version> - fnm:
fnm install <version>thenfnm use <version> - Other: https://nodejs.org (LTS)
- Forge CLI — Run
forge --version. If missing:
npm install -g @forge/cli
- Forge login — Run
forge whoami. If not logged in:
- Never ask for or accept API tokens in chat — tokens are sensitive; the user must enter them only in their terminal
- Direct the user to create an API token: https://id.atlassian.com/manage/api-tokens
- Tell the user to run
forge loginin their own terminal (not via the agent). The CLI will prompt for:- Atlassian email
- API token (paste only in the terminal when prompted)
- Example message: "You need to log in to Forge. Create an API token at https://id.atlassian.com/manage/api-tokens, then run
forge loginin your terminal. Enter your email and token when prompted — do not paste them here." - After the user confirms they've logged in, retry the workflow
Install in order: Node.js first (required for npm), then Forge CLI, then login. Retry the workflow after installing.
Step 1: Discover Developer Spaces
forge developer-spaces list --json
Step 2: Ask User to Choose Developer Space
Do not proceed to Step 3 until the user has selected a developer space. Present the list from Step 1 (names and IDs) and ask which space to use. If only one space exists, you may use it and briefly inform the user. Never pick one of multiple spaces on the user's behalf.
Step 3: Create App
All python3 -m scripts.* commands must be run from the skill directory (the directory containing this SKILL.md file). Derive it from the SKILL.md path provided in the system prompt. Use python3 on macOS if python is not available.
Run from the skill directory. The --directory flag sets the parent directory under which the app folder (named after --name) will be created. The script cds into that directory before running forge create, so the app appears as a subdirectory (e.g. <parent-directory>/<app-name>/). If omitted, the app is created under the current directory.
python3 -m scripts.create_forge_app \
--template <template> \
--name <app-name> \
--dev-space-id <selected-id> \
--directory <parent-directory>
To find the right template for the user's needs:
- Call
list-forge-modulesto identify the appropriate module type - Call
search-forge-docswith a query like "template for " to find the matching template name
Validate a template: python3 -m scripts.list_templates --validate <name>
List all templates: python3 -m scripts.list_templates --list
Step 4: Customize Code
After forge create succeeds:
cd <app-name>
npm install
UI Kit vs Custom UI — Choose the Right Tools
Before writing any UI code, determine which approach the app uses. Getting this wrong causes import errors and broken builds.
- UI Kit (most
forge createtemplates): manifest usesrender: nativeor code imports from@forge/react. Useforge-ui-kit-developer-guideas the ONLY UI reference. Do NOT suggest@atlaskit/* imports — they won't work. - Custom UI: manifest has
resourcepointing to astatic/directory. Use ADS MCP tools (ads_plan,ads_get_components,ads_get_all_icons) for component discovery. Do NOT useforge-ui-kit-developer-guide— it describes a different API.
Knowledge tools for implementation
forge-ui-kit-developer-guide— Frontend components (UI Kit only)ads_plan/ads_get_components— Component and token lookup (Custom UI only)forge-backend-developer-guide— Backend resolvers and APIsforge-app-manifest-guide— Manifest configurationsearch-forge-docs— Search for specific APIs or props
Step 5: Deploy and Install (run the deploy script)
You MUST run the deploy script — do not only paste manual forge deploy / forge install commands for the user to run. Execute the script from the skill directory.
- If you have the user's Atlassian site URL (e.g. they provided it earlier or in the request), run in one go:
python3 -m scripts.deploy_forge_app \
--app-dir <app-directory> \
--site <site-url> \
--product <jira|confluence>
- If you do not have the site URL yet: run deploy only, then ask for the site, then run again to install:
# 1) Deploy only
python3 -m scripts.deploy_forge_app \
--app-dir <app-directory> \
--product <jira|confluence> \
--deploy-only
# 2) Ask the user: "What is your Atlassian site URL (e.g. yourcompany.atlassian.net)?"
# 3) After they reply, run again with their site to complete install
python3 -m scripts.deploy_forge_app \
--app-dir <app-directory> \
--site <site-url> \
--product <jira|confluence> \
--skip-deps
Always ask the user for their site URL when needed for install; never try to discover it. If scopes changed from a previous install, add --upgrade to the install (or run the script again with --site and the script will handle it; for manual upgrade use forge install ... --upgrade).
Cross-product installation
When an app uses scopes from multiple products (e.g. a C