惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博客园 - 聂微东
Y
Y Combinator Blog
WordPress大学
WordPress大学
L
LangChain Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
A
About on SuperTechFans
小众软件
小众软件
有赞技术团队
有赞技术团队
S
SegmentFault 最新的问题
宝玉的分享
宝玉的分享
Recent Announcements
Recent Announcements
GbyAI
GbyAI
I
InfoQ
The GitHub Blog
The GitHub Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
酷 壳 – CoolShell
酷 壳 – CoolShell
罗磊的独立博客
C
Check Point Blog
V
V2EX
Apple Machine Learning Research
Apple Machine Learning Research
月光博客
月光博客
量子位
雷峰网
雷峰网
Hugging Face - Blog
Hugging Face - Blog

Show HN

GitHub - astefanutti/shaderbang: Shebang for Shaders Show HN: Generate Claude Code Workflows using Spec Driven Development approach Show HN: AI agents for UK GDAD PCF roles and their skills The Two Pillars: Mixer Mode and Meta-Software in the Reorganization of Software Work After AI GitHub - JaiCode08/teleport-env What 1,000+ Harness Experiments Taught Me About Self-Improving Agents Show HN: Liiists, a Markdown-first, iOS and CLI list app SwiperTab – Get this Extension for 🦊 Firefox (en-US) GitHub - kouhxp/fftext: Summarize, explain, fact-check, or translate any text, URL, or file. No GPU. No cloud. One command GitHub - sweetpad-dev/sweetpad: Develop Swift/iOS projects using VSCode GitHub - dogmaticdev/IRON: IRON a.k.a. Intermediate Representation Object Notation is a Interpreter/Database that is used to create Programming Languages. GitHub - sjhalani7/vaen: Package your AI coding harness into a portable .agent file, and share it across repos, teams, & the community without ever having to copy-paste instructions, skills, MCP config, or secrets. Show HN: Gandalf the Grader Show HN: Citadeld – replay any CI failure locally from a single file GitHub - tdortman/cuSBF: High-Performance GPU Super Bloom Filter coral-ai/claude-code-token-xray at main · Coral-Bricks-AI/coral-ai GitHub - ulyssestenn/funes: Funes is a Git-based framework for LLM-managed knowledge work: an AI Librarian ingests raw sources, builds an interlinked Markdown knowledge base, and uses it to produce cited reports, analyses, and other outputs. GitHub - ThatXliner/gah: Git Add Hunk, built for agents to use GitHub - harmont-dev/harmont-cli: Command-line client for the Harmont CI platform GitHub - brooksmcmillin/mcp-authflow: OAuth 2.0 Authorization Server framework for MCP servers GitHub - javaid-codes/audit-supply-chain-agents GitHub - amorey/gochan: A small library of common channel architectures for Go, inspired by Rust GitHub - arifozgun/OpenGem: Free, Open-Source AI API Gateway with Gemini, OpenAI & Anthropic Compatibility in 1 file GitHub - Pranesh950/BioPetals: 🌸 Run BIOxAI models at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading GitHub - cnguyen14/bounty-doctor: Diagnose a GitHub bounty issue before you waste hours: detects honeypot scam repos, AI-bot attempt swarms, and stale contests. Show HN: CoreMCP – MCP Server for On-Prem DBs Show HN: KittyHTML – Render HTML/CSS as an inline image in your terminal GitHub - bingud/filemat: Web-based file manager Show HN: TruthLens – Free multi-signal deepfake image detector GitHub - apexlocal-jz/claude-usage-tray: Windows system-tray app showing your Claude Code rate-limit usage at a glance. Zero deps, ~300 lines of PowerShell. Cross-IDE (works regardless of VS Code, Cursor, plain terminal).
GitHub - dedalus-labs/hollywood: Lights, Cameras, (GitHub...
windsor · 2026-06-17 · via Show HN

Lights, cameras, Actions!

Hollywood lets you write GitHub Actions logic as typed TypeScript, run it locally, and generate ordinary GitHub Actions files for CI/CD.

"Lights, Cameras, (GitHub) Actions!"

Hollywood is AI-native and AI-friendly. The docs ship copy-page controls and generated agent context. Point your agents at the docs, hand them llms.txt or llms-full.txt, and let them rip on typed TypeScript actions instead of hand-writing YAML.

GitHub Actions is a good orchestration layer. It knows when jobs should run, which runner labels they need, which secrets exist, and how jobs depend on each other.

It is a rough programming environment. Real DevOps logic often turns into shell inside YAML: untyped strings, quoting bugs, hidden input coercion, and commits whose only purpose is "try CI again".

Our position is simple: people should not spend their time painstakingly handwriting imperative GitHub Actions YAML. YAML should orchestrate. TypeScript should program.

Hollywood moves the imperative part into TypeScript scripts you can test before they run on GitHub. The generated output is still boring GitHub Actions: action.yml, uses: ./.github/actions/..., and JavaScript actions that run through GitHub's official action toolkit.

This works because GitHub Actions can run JavaScript actions directly. An action.yml file points at a Node entrypoint, and Hollywood generates the thin adapter around your typed script.

See CONTRIBUTING.md for the CLA/Vouch contribution flow and ROADMAP.md for planned contribution areas. See SECURITY.md for the GitHub Actions hardening policy.

Docs

Published docs live at https://oss.dedaluslabs.ai/hollywood.

Build them locally:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r docs/requirements.txt
python -m mkdocs serve -f mkdocs.yml

Contributions

Hollywood accepts external code from vouched contributors. Due to the increased volume of AI-generated code, Hollywood uses Vouch as the arbiter of contributor trust and CLA eligibility for external pull requests. Being listed in VOUCHED.td means a maintainer has verified the GitHub account and recorded that the contributor accepted CLA.md.

The flow is:

  1. Open a "Vouch request" issue.
  2. Confirm that you have read and accept CLA.md.
  3. Link public GitHub work, a project website, or another public identity that helps a maintainer recognize you.
  4. If an existing vouched contributor knows you, ask them to comment on the issue.
  5. A maintainer adds your GitHub handle to VOUCHED.td.

Do not add yourself to VOUCHED.td in your first contribution. The CLA check reads that file from the trusted base branch, so normal pull requests cannot self-vouch.

For code and docs changes, fork the repository and open a pull request from your branch into dedalus-labs/hollywood:main. See CONTRIBUTING.md for the full checklist.

Node Requirements

The package runtime and the repository toolchain have different Node requirements:

Surface Node requirement
Installed package and CLI Node 20 or newer
Generated GitHub actions GitHub's Node 24 action runtime
Building Hollywood locally Node 22.18+ or Node 24.11+

The published package declares engines.node >=20 in package.json. The build output targets Node 20 in tsdown.config.ts. tsconfig.json is only the typecheck configuration; it is not the runtime support contract.

Use Node 22.18+ or Node 24.11+ when contributing because the local build and declaration-generation toolchain has stricter engine requirements than the published runtime package.

Install

npm install --save-dev @dedalus-labs/hollywood

That installs a local hollywood binary at node_modules/.bin/hollywood. Run it with npx hollywood ..., or put hollywood ... inside an npm script.

Small Dependency Surface

Hollywood is intentionally lightweight. The package has six direct runtime dependencies:

  • @actions/core
  • @actions/exec
  • @actions/expressions
  • @actions/workflow-parser
  • esbuild
  • yaml

Most of that surface is GitHub's own action toolkit and schema parser. The published package only ships runtime files, type declarations, package metadata, the README, and the license. A smaller dependency graph is easier to audit and reduces npm supply-chain exposure.

Before / After

Before Hollywood, a container publish step might look like this:

- name: Publish container image
  run: |
    set -euo pipefail
    IMAGE_REF="ghcr.io/acme/api:${GITHUB_SHA}"
    docker buildx build \
      --file Dockerfile \
      --tag "${IMAGE_REF}" \
      --push \
      --provenance false \
      .
    echo "image_ref=${IMAGE_REF}" >> "$GITHUB_OUTPUT"

With Hollywood, the program is typed TypeScript instead of text hidden in YAML:

import {
	type ActionInputValues,
	type ActionOutputValues,
	action,
	booleanInput,
	choiceInput,
	integerInput,
	pathInput,
	stringInput,
	stringOutput,
} from "@dedalus-labs/hollywood";

const publishInputs = {
	image: stringInput({ description: "Container image name, including registry." }),
	tag: stringInput({ description: "Container image tag." }),
	context: pathInput({ description: "Build context path.", default: "." }),
	dockerfile: pathInput({ description: "Dockerfile path.", default: "Dockerfile" }),
	platform: choiceInput({
		description: "Build target platform.",
		options: ["linux/amd64", "linux/arm64"] as const,
		default: "linux/amd64",
	}),
	provenance: choiceInput({
		description: "Build provenance mode.",
		options: ["false", "min", "max"] as const,
		default: "false",
	}),
	cacheFrom: stringInput({ description: "Optional build cache source.", default: "" }),
	buildAttempt: integerInput({ description: "CI build attempt number." }),
	push: booleanInput({ description: "Push instead of loading locally.", default: "true" }),
} as const;

const publishOutputs = {
	imageRef: stringOutput({ description: "Published image reference." }),
} as const;

type PublishImageInput = ActionInputValues<typeof publishInputs>;
type PublishImageOutput = ActionOutputValues<typeof publishOutputs>;

const imageRef = (input: Pick<PublishImageInput, "image" | "tag">): string =>
	`${input.image}:${input.tag}`;

const dockerBuildArgs = (input: PublishImageInput, ref: string): readonly string[] => {
	const args = [
		"buildx",
		"build",
		"--file",
		input.dockerfile,
		"--platform",
		input.platform,
		"--tag",
		ref,
		"--label",
		`ci.build-attempt=${input.buildAttempt}`,
		"--provenance",
		input.provenance,
	] as string[];

	if (input.cacheFrom.length > 0) {
		args.push("--cache-from", input.cacheFrom);
	}
	args.push(input.push ? "--push" : "--load", input.context);
	return args;
};

export const publishImage = action({
	name: "publish-container-image",
	description: "Build and publish a container image without embedding shell in workflow YAML.",
	inputs: publishInputs,
	outputs: publishOutputs,
	run: async ({ exec, input }): Promise<PublishImageOutput> => {
		const ref = imageRef(input);
		await exec("docker", dockerBuildArgs(input, ref));
		return { imageRef: ref };
	},
});

Hollywood parses GitHub's string inputs into PublishImageInput before run starts. You can still layer Zod, Effect Schema, or your own parser on top for repository-specific policy:

import { z } from "zod";

const publishPolicy = z.object({
	image: z.string().regex(/^ghcr\.io\/[a-z0-9-]+\/[a-z0-9._/-]+$/),
	tag: z.string().min(1).max(128).regex(/^[A-Za-z0-9_.-]+$/),
	context: z.string().refine((path) => !path.includes(".."), "context must stay inside workspace"),
	push: z.boolean(),
});

const validatePublishPolicy = (input: PublishImageInput): void => {
	publishPolicy.parse(input);
};

export const publishImage = action({
	// ...
	run: async ({ exec, input }): Promise<PublishImageOutput> => {
		validatePublishPolicy(input);
		const ref = imageRef(input);
		await exec("docker", dockerBuildArgs(input, ref));
		return { imageRef: ref };
	},
});

Those schema packages live in your workflow repository. Hollywood does not pull them into its own runtime dependency graph.

GitHub still sees a normal local action step:

- name: Publish container image
  uses: ./.github/actions/publish-container-image
  with:
    image: ghcr.io/acme/api
    tag: ${{ github.sha }}
    context: .
    dockerfile: Dockerfile
    platform: linux/amd64
    provenance: "false"
    build-attempt: ${{ github.run_attempt }}
    push: "true"

The important bit is the command shape:

const args = [
	"buildx",
	"build",
	"--file",
	input.dockerfile,
	"--platform",
	input.platform,
	"--tag",
	ref,
	"--label",
	`ci.build-attempt=${input.buildAttempt}`,
	input.context,
];

await exec("docker", args);

That is execve(2)-shaped: one executable path and one array of arguments. There is no shell interpolation and no YAML quoting puzzle.

Local Runs

Run an exported action directly on your machine:

npx hollywood run gha/containers/publish-image.ts \
  --export publishImage \
  --with image=ghcr.io/acme/api \
  --with tag="$(git rev-parse --short HEAD)" \
  --with context=. \
  --with dockerfile=Dockerfile \
  --with buildAttempt=1 \
  --with provenance=false

Route every exec(file, args) call through a Lima VM when the script needs a Linux environment:

npx hollywood run gha/cache/s3-cache.ts \
  --export s3Cache \
  --lima default \
  --start-vm \
  --with mode=restore \
  --with bucket=ci-cache \
  --with prefix=node \
  --with key=linux-arm64 \
  --with archivePath=/tmp/cache.tar.gz \
  --with contentsPath=/tmp/node-cache

Hollywood invokes Lima with the same argument-array shape:

limactl shell --tty=false --start default -- <file> <arg>...

No command is rewritten into shell text. If the VM is stopped and --start-vm was not passed, the run fails before the action starts. See the execution backend docs for the supported Lima backend and planned backend directions.

Generate Actions

Generate local action metadata and entrypoints:

npx hollywood generate "gha/**/*.ts" --output .

Hollywood writes ordinary GitHub Actions files:

.github/actions/publish-container-image/action.yml
.github/actions/publish-container-image/src/index.ts
.github/workflows/container-release.yml

Generated files include a marker:

# @generated by Hollywood. Do not edit by hand.

Edit the TypeScript source and regenerate. We recommend not hand-patching generated YAML.

Workflow Sources

Hollywood can generate workflow YAML from typed workflow objects too:

import { generateWorkflowFile, job, uses, workflow } from "@dedalus-labs/hollywood";
import { gh } from "@dedalus-labs/hollywood/expr";
import { publishImage } from "./containers/publish-image";

export const containerRelease = workflow({
	name: "Container Release",
	on: {
		push: { branches: ["main"] },
		workflow_dispatch: {},
	},
	permissions: { contents: "read", packages: "write" },
	jobs: {
		publish_image: job({
			"runs-on": "ubuntu-latest",
			steps: [
				{ uses: "actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10" },
				uses(publishImage, {
					name: "Publish container image",
					with: {
						image: "ghcr.io/acme/api",
						tag: gh.github.sha,
						provenance: "false",
					},
				}),
			],
		}),
	},
});

export default generateWorkflowFile({
	sourcePath: "gha/container-release.ts",
	sourceRoot: "gha",
	workflowsDir: ".github/workflows",
	workflow: containerRelease,
});

Use Cases

Hollywood is useful when the CI/CD step is a real program:

  • publishing container images
  • creating release artifacts
  • promoting GitOps manifests between environments
  • running Terraform plan/apply wrappers
  • restoring and saving object-storage-backed caches
  • validating pull requests with path-dependent jobs

Hollywood is not a local GitHub Actions emulator. GitHub still decides event payloads, runner labels, secrets, permissions, and job scheduling.

Roadmap

Future work is tracked in ROADMAP.md. Concrete tasks should become GitHub issues before implementation, especially if they change the public API or generated YAML.

LICENSE

MIT.

Development

npm ci
npm test
npm run build
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r docs/requirements.txt
python -m mkdocs build --strict -f mkdocs.yml