Generate Manifest logo

Generate Manifest

OrganizationPopular
google
generate-manifest

Scan an ADK recipe directory and generate a manifest.yaml for it based on the schema at .github/schemas/manifest-schema.json. Use when the user wants to create or generate a manifest.yaml for a recipe under core/, contrib/, or plugins/.

Overview

Publishergoogle
Repositoryadk-recipes
Skill namegenerate-manifest
Stars
10.4K
Forks
2.9K
Bundled files
Instructions only
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by google on GitHub. Read the source before you install it.

Installation

Install the Generate Manifest AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/google/adk-recipes/tree/main/.agents/skills/generate-manifest
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/google/adk-recipes.git /tmp/adk-recipes
mkdir -p .claude/skills
cp -r /tmp/adk-recipes/.agents/skills/generate-manifest .claude/skills/generate-manifest
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Generate Manifest in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Generate Manifest on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Generate Manifest is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Generate Manifest

Scan a recipe directory and produce a valid manifest.yaml for it, then validate it against the schema.

Process

1. Identify the target recipe

The user will provide a path to a recipe directory (e.g. core/python/rag-agent-search, contrib/python/my-recipe, plugins/retail/store-ops). If they don't, ask for it before proceeding.

2. Read the schema

Read .github/schemas/manifest-schema.json from the repo root to understand the current required and optional fields, allowed enum values, and constraints. Do not rely on memory — always read the live schema file so changes to it are automatically picked up.

3. Scan the recipe directory and infer only what the code proves

Read the following files if they exist:

  • pyproject.toml / requirements.txt — primary language (for language field)
  • app/agent.py or equivalent entry point — single vs multi-agent, agent count
  • Makefile — deploy targets (determines deployable)
  • app/ source files — stateful writes, datasource patterns
  • AGENTS.md — architecture notes if present

Field-by-field rules — read carefully before filling any field:

FieldRule
typeInfer from code. standalone = has its own runnable entry point: look for an if __name__ == "__main__" block, a make playground/run target, an adk web/adk run invocation, or a CLI. module = only importable (it just exports a root_agent/Agent for another workflow to orchestrate, with no way to run on its own). When ambiguous, re-read the schema's type description and prefer module only when there is genuinely no entry point.
deployableOPTIONAL. true only if a single make target or script deploys everything with no manual steps. Omit the field entirely if false (schema default is false).
largeOPTIONAL. Whether to opt into the relaxed size tier. Exact limits depend on whether the recipe lives under core/ or contrib/ or plugins/ and are defined in .github/policy.yml (recipe_size_limits) — as of writing: contrib/ and plugins/ default is 70 files / 2 MB and large is 200 files / 10 MB; core/ default is 500 files / 50 MB. Omit the field unless the recipe would otherwise exceed the default tier's limits (schema default is false).
statusAlways "active" unless there is explicit evidence of abandonment.
languageRead from file extensions or pyproject.toml. Never guess.
descriptionRead README.md and AGENTS.md only (author-written intent, not code). Write a draft of at most 15 words summarising what the recipe does. The schema requires at least 10 characters, so keep the draft comfortably above that. Append the comment # TODO: review and expand this draft description. If neither file exists or the intent is unclear, fall back to "DESCRIPTION" with the same TODO comment.
architecture.agentInfer from code only: count Agent( or equivalent constructor calls. single or multi. Omit the whole architecture block if uncertain.
architecture.statefulInfer from code: true only if the recipe writes persistently to an external system (DB, vector store, GCS) during normal operation — not just one-time setup.
architecture.datasourcesInfer from code: hardcoded = data literals in source; local = files bundled in repo; external = live systems queried at runtime. Can be multiple.
dependenciesInclude the block but comment it out entirely. Do NOT infer library or service names — GCP product names change and guesses will be wrong. Use the commented-out sample shown in the template.
ownership.teamAlways use the exact placeholder "TODO: Replace with your team name" — this is the literal string the validator checks for, so it correctly fails validation until a human fills it in. Never invent or infer a team name.
ownership.pocAlways use the exact placeholder "TODO: Replace with your GitHub user ID" — this is the literal string the validator checks for. Never invent or infer a GitHub ID from email addresses, file authors, or any other source.
ownership.contributorsOPTIONAL. Omit the field entirely. Leave a commented-out sample line so the author knows it exists. Never infer contributor IDs.
tagsInclude but comment out entirely, with a sample entry showing the expected style. Do NOT generate real tags — tag choices are the author's call.
licenseOPTIONAL. If the recipe's source tree contains a LICENSE file or the author has explicitly stated a license, record the SPDX identifier (e.g. "Apache-2.0", "MIT", "BSD-3-Clause"). Otherwise comment it out. Never infer or guess — only set it if it is explicitly declared.

4. Make judgment calls explicitly

Before writing the manifest, briefly state what you found for each inferred field (type, deployable, architecture.*) and cite the specific file/line that supports your conclusion. This lets the user catch errors before they land in the file.

5. Write the manifest

Write manifest.yaml to the root of the recipe directory. Match the concise inline-comment style used by the existing manifests in core/ and contrib/ — allowed values shown as # Options: [...] after each field:

yaml
type: "..."          # Options: [standalone | module]
status: "active"     # Options: [active | inactive]
language: "..."      # Options: [python | java | go | kotlin | typescript]
description: "..."   # TODO: review and expand this draft description

# deployable: true   # (optional) one-command deploy, no manual steps; omit if false (default)
# large: true        # (optional) opt into the relaxed size tier; see .github/policy.yml (recipe_size_limits) for the exact numbers; omit if false (default)

architecture:          # (optional) omit the whole block if nothing below can be inferred from code
  agent: "..."          # Options: [single | multi]
  stateful: false       # Options: [true | false]
  datasources:          # Options: [hardcoded | local | external]
    - "..."

# dependencies: (optional) uncomment and fill in with canonical names
#   libraries:
#     - "ADK"
#     - "pandas"            # example — replace with actual libraries used
#   services:
#     - "GCP Project"
#     - "Cloud Run"         # example — replace with actual GCP/external services used

ownership:
  team: "TODO: Replace with your team name"
  poc: "TODO: Replace with your GitHub user ID"
  # contributors: (optional) uncomment and add GitHub IDs of additional contributors
  #   - "github-id-1"

# tags: (optional) uncomment and replace with meaningful labels
#   - "rag"               # example — use technology names, patterns, and use-case keywords
#   - "gemini"

# license: "Apache-2.0"  # (optional) SPDX identifier; only set if explicitly declared — see https://spdx.org/licenses/

Omit architecture entirely if none of its sub-fields can be determined from the code.

6. Validate

After writing the manifest, run the validator from the repo root (do not cd into tools/, and do not pass --active):

bash
uv run validate manifest <recipe-path>

<recipe-path> must be relative to the repo root (e.g. core/python/my-recipe, contrib/python/my-recipe, plugins/retail/store-ops), not an absolute path.

Interpret the result carefully — there are two kinds of failure:

  • Real schema errors (missing required field, invalid enum value, description under 10 characters, malformed YAML): fix these in the manifest and re-validate until they are gone.
  • ownership.team / ownership.poc placeholder errors: these are EXPECTED and must be left as-is. The validator deliberately fails while the placeholders are in place, to force a human to supply real values before merge. Do NOT invent a team name or GitHub ID to make validation pass.

So a correctly generated manifest will still report [FAIL] — but only on the two ownership placeholders, and nothing else. That is the intended outcome; report it to the user as "valid except for the ownership placeholders you need to fill in". Do not change or modify any other files in the repository. Also do not commit the changes.

7. Report back

Tell the user:

  1. What was inferred from the code and which file/line supports each inference.
  2. Which fields need human input before the manifest will pass validation: ownership.team and ownership.poc (placeholders — validation intentionally fails until these are replaced with real values). Also flag description (draft generated — must be reviewed and expanded) and dependencies/tags (commented out, ready to uncomment and fill in).

Frequently asked questions

What does the Generate Manifest AI skill do?

Scan an ADK recipe directory and generate a manifest.yaml for it based on the schema at .github/schemas/manifest-schema.json. Use when the user wants to create or generate a manifest.yaml for a recipe under core/, contrib/, or plugins/.

Why use Generate Manifest on TypingMind?

Because you install it once and use it with any model. Generate Manifest is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Generate Manifest in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/google/adk-recipes/tree/main/.agents/skills/generate-manifest. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Generate Manifest?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Generate Manifest?

As many as you like. As long as a model supports skills, you can use Generate Manifest with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Generate Manifest AI skill free?

Yes. It is published on GitHub by google under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇