Ffcpe Custom Node Sdk

FFCPE custom actions on App Builder (@adobe/ffcpe-custom-node-core, @adobe/ffcpe-custom-node-app-builder)

Audience

Use this skill when assisting consumers of the NPM packages: teams shipping an App Builder application that Workflow Builder invokes as a custom-action (async submit + status polling).

For contributing to the SDK source repository (monorepo, Turbo, releases), see README.md (“Monorepo (contributors)”) and CONTRIBUTING.md next to the SDK sources (internal mirror or published repo checkout).

For scaffolding web/worker files, webpack, and ext.config.yaml, use ffcpe-app-builder-actions.

For catalog-entry.json authoring and aio ffcpe catalog (validate, register, list, update), install skills from adobe/aio-cli-plugin-ffcpe: ffcpe-catalog-entry-json and aio-ffcpe-cli.

npx skills add adobe/aio-cli-plugin-ffcpe --all -y

End-to-end checklist

0. Bootstrap Console project (new projects only) — create project + workspace via aio console, run aio app init --standalone-app, download workspace config, wire with aio app use <config>.json --overwrite --no-input. Clean up web-src/, scaffolded generic actions, and web: web-src from app.config.yaml. Replace webpack-config.js with the esbuild-loader + commonjs2 shape. Console project names: alphanumeric only, ≤20 chars (no hyphens); may differ from the OpenWhisk package name in YAML. See ffcpe-app-builder-actions → "Starting from scratch with init-bare" for the exact sequence.

1. Install SDK packages (npm or pnpm) and implement web + worker actions. Best practice: co-locate the catalog entry as <action-name>.entry.json next to <action-name>.web.* and <action-name>.worker.* (see ffcpe-app-builder-actions).

2. Deploy the App Builder app; copy web action base URLs from deploy output; set catalog submitEndpoint / statusEndpoint to {base}/submit and {base}/status.

3. Author <action-name>.entry.json — ports must match worker input/output names; endpoints from deploy; authentication: { "type": "ims_service_token" } when using default mountFfcpeNodeRoutes auth (not none unless authenticate: null on routes).

4. Validate and register with the CLI — never raw curl:

   aio ffcpe catalog validate --file ./actions/my-action/my-action.entry.json
   aio ffcpe catalog register --file ./actions/my-action/my-action.entry.json
   

Install packages

with npm:

npm install @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder

or pnpm — set auto-install-peers=true in the app’s .npmrc, then:

pnpm add @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder

Explicit peers — if you prefer every peer in package.json, or your tool does not auto-install peers (e.g. Yarn Classic):

npm install @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder hono hono-openwhisk-adapter

pnpm add @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder hono hono-openwhisk-adapter

Typical peers (confirm versions against your template and Adobe docs):

• @adobe/aio-lib-core-logging, @adobe/aio-lib-state, @adobe/aio-lib-ims, openwhisk, hono, hono-openwhisk-adapter

Link a local SDK checkout

Use this when the consumer app should load @adobe/ffcpe-custom-node-core / @adobe/ffcpe-custom-node-app-builder from a nearby clone of the SDK monorepo instead of the registry.

1. In the SDK monorepo, install and build dist/ for both packages:

   pnpm install && pnpm run build
   # or: npm install && npm run build
   

   For tight iteration, run pnpm run dev (or npm run dev) from the SDK root so dist/ rebuilds on change.

2. In the consumer App Builder project, link core first, then app-builder:

   pnpm:

   pnpm link /path/to/ffcpe-custom-node-sdk/packages/core
   pnpm link /path/to/ffcpe-custom-node-sdk/packages/app-builder
   

   npm:

   npm link /path/to/ffcpe-custom-node-sdk/packages/core
   npm link /path/to/ffcpe-custom-node-sdk/packages/app-builder
   

3. To restore registry packages:

   pnpm unlink @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder && pnpm install
   # or: npm unlink @adobe/ffcpe-custom-node-core @adobe/ffcpe-custom-node-app-builder && npm install
   

Optional global link (pnpm): register each package with pnpm link --global from packages/core and packages/app-builder, then in the consumer pnpm link --global @adobe/ffcpe-custom-node-core and pnpm link --global @adobe/ffcpe-custom-node-app-builder. If workspace:* fails to resolve for app-builder, keep using directory links or add a consumer pnpm.overrides pinning @adobe/ffcpe-custom-node-core to the packages/core path.

Full wording lives in the SDK README.md (“Use a local SDK checkout”).

Install agent skills

Skills live in the SDK repo under skills/ and install via the open agent skills CLI. One canonical copy is symlinked into each agent (Claude Code, Cursor, etc.) so guidance stays identical.

From GitHub (consumer repo)

# List available skills
npx skills add adobe/ffcpe-custom-node-sdk --list

# Install all FFCPE skills (project-local)
npx skills add adobe/ffcpe-custom-node-sdk --all -y

# Install only this skill
npx skills add adobe/ffcpe-custom-node-sdk --skill ffcpe-custom-node-sdk -y

# Claude Code + Cursor
npx skills add adobe/ffcpe-custom-node-sdk --all -a claude-code -a cursor -y

# Global (all projects on this machine)
npx skills add adobe/ffcpe-custom-node-sdk --all -g -y

From a local SDK checkout

npx skills add /path/to/ffcpe-custom-node-sdk/skills --all -y

See skills/README.md in this repo for update/remove commands and skill layout.

CLI plugin skills (separate repo):

npx skills add adobe/aio-cli-plugin-ffcpe --all -y

Reference skills by name in each SKILL.md frontmatter: ffcpe-custom-node-sdk, ffcpe-app-builder-actions, aio-ffcpe-cli, ffcpe-catalog-entry-json.

Catalog CLI (@adobe/aio-cli-plugin-ffcpe)

Install the Adobe I/O CLI plugin for catalog operations after deploy:

npm install -g @adobe/aio-cli
aio plugins:install @adobe/aio-cli-plugin-ffcpe
aio login
aio console org select

Use --strict on register/update to fail on validation warnings. Override API host with AIO_FFCPE_CATALOG_BASE_URL or --base-url. Prefer aio ffcpe catalog over hand-written HTTP — auth uses the same IMS session as aio login.

Align customActionConfig.submitEndpoint / statusEndpoint in catalog JSON with your deployed web action URLs and route paths from mountFfcpeNodeRoutes. Port name values in catalog inputs / outputs must match what the worker reads via getTextInput, getImageInput, etc.

Two OpenWhisk actions (recommended shape)

1. Web action — HTTP entry for FFCPE: POST /submit, GET /status. Build a Hono app, call mountFfcpeNodeRoutes, export main with ToOpenWhiskAction from hono-openwhisk-adapter.
2. Worker action — Long-running job: export main from createFfcpeNodeWorker(handler, options).

Use ext.config.yaml (or equivalent) so the web action uses web: "raw" and annotations.require-adobe-auth: false (see README.md manifest example). Match worker / web package and action names to what mountFfcpeNodeRoutes receives.

Web route wiring (mountFfcpeNodeRoutes)

• Pass worker: { package, name } — target for async OpenWhisk invoke after submit.
• Pass web: { package, name } — required; used to build statusUrl returned to Workflow Builder. Omitting or empty strings throws at mount time.
• Optional routes — defaults /submit and /status; align with catalog submitEndpoint / statusEndpoint paths.
• Optional authenticate — defaults to IMS bearer + x-api-key via createImsInboundAuth. Pass null to disable auth on those routes (only if product/security allows).
• Optional jobStore, loggerName, jobMetadata(submit) — advanced; default store is aio-lib-state-backed.

Worker handler (createFfcpeNodeWorker)

• Implement FfcpeNodeWorkerHandler: async ({ jobId, inputs, params, logger, authContext, args }) — inputs is the validated FFCPE input[]; params is workflow parameters from run-workflow (submit parameters); args is the full worker invocation object (OpenWhisk main(args)).
• Return { status: "completed", outputs?: ActionOutput[] } or { status: "failed", error: string }.
• Use requiredInputNames in options to fail fast when submit payloads omit named inputs.
• Throw FfcpeJobFailedError only when intentionally mapping failures to the SDK’s failed-job behavior (the adapter already maps handler failed results).

Prefer findInputByName, getTextInput, etc., from @adobe/ffcpe-custom-node-core, and narrow with TextInput, ImageInput, VideoInput, JsonInput. Build outputs with createTextOutput, createImageOutput, createVideoOutput, createJsonOutput (and related helpers).

HTTP contract (what Workflow Builder expects)

Workflow Builder POSTs { inputs, parameters } to submitEndpoint and polls status until completed or failed. Submit responses should include jobId, status, and ideally statusUrl for polling.

For header rules, response fields, and polling semantics, read docs/custom-action-requests.md from the SDK documentation bundle or source tree (same material ships beside the repo README for maintainers).

Security and robustness (custom-action authors)

• Treat inputs and parameters as untrusted; validate types, sizes, and URLs before fetching or processing assets.
• Never log tokens, API keys, or raw Authorization headers; keep Logger output minimal.
• Use HTTPS asset URLs where possible; fail closed on suspicious schemes or redirects if implementing custom fetch logic outside the SDK.
• Keep auth enabled on submit/status unless there is an explicit, reviewed reason to pass authenticate: null.

Advanced / escape hatches

• Core-only HTTP — handleFfcpeSubmit, handleFfcpeStatus, or runWorkerJob when not using Hono (still need JobStore, JobOrchestrator, InboundAuth, status URL builder).
• Custom orchestration or storage — implement core ports (JobStore, JobOrchestrator) and mirror the checklist in the upstream README section “Implementing for a different Cloud Functions Provider” (same ideas apply when forking adapter behavior).

Quick lookup

For import cheat sheets, pnpm link, and troubleshooting (status URL, web manifest), see references/quick-reference.md.