google / google/adk-python-community

Add standardized planner content blocks helper (community module)

Ouverte
#162 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub
Langage dominant
Python
Étoiles
182
Forks
75
Métriques de merge des PR
Aucune PR mergée en 30 j

Description

### What

Add a small, read-only helper that converts planner-produced `google.genai.types.Part` lists into provider-agnostic, typed "content blocks" modelled after [LangChain v1's standard content blocks](https://docs.langchain.com/oss/python/langchain/messages#standard-content-blocks).

ADK planners (`PlanReActPlanner`, `BuiltInPlanner`) emit reasoning as parts with `thought=True`, and `PlanReActPlanner` additionally annotates text with inline `/*PLANNING*/`-style tags that consumers must parse by hand. This helper turns that into a list of typed blocks — `reasoning` (with a `reasoning_kind` of planning/replanning/reasoning/action), `text`, and `tool_call` — so consumers branch on a `type` discriminator instead of re-parsing raw text.

It is **additive and read-only**: it never mutates the parts and doesn't change how planners build instructions or post-process responses. Existing consumers of the `Part` output are unaffected.

### Background

Originally proposed in core ADK (google/adk-python#6185). That PR was closed and the maintainer suggested the feature belongs in `adk-python-community` as a community module. This issue tracks landing it here.

### Proposed location

New module `src/google/adk_community/planners/` exporting `parts_to_content_blocks` / `part_to_content_block` (plus the `ReasoningContentBlock` / `TextContentBlock` / `ToolCallContentBlock` TypedDicts, `ReasoningKind`, and `ContentBlock`). Self-contained — depends only on `google-genai` (already a transitive dependency via `google-adk`), so no new optional-dependency group is needed. Unit tests under `tests/unittests/planners/`.

A PR implementing this is ready to open.

Guide de contribution

Ouvrir le guide de contribution

Piste de recherche

Commencez par le chemin de module proposé src/google/adk_community/planners/ et examinez google.genai.types.Part ainsi que les blocs de contenu standard de LangChain v1. Implémentez les helpers de conversion en lecture seule et les types TypedDict décrits dans l’issue, puis ajoutez des tests unitaires sous tests/unittests/planners/. Le travail est terminé lorsque les blocs reasoning, text et tool_call exposent les discriminateurs de type et les types de reasoning indiqués, sans modifier les parts d’entrée.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Évaluation

Stack technique
python
Domaine
ai
Type d'issue
Fonctionnalité
Difficulté
3/5
Temps estimé
1-2 jours
Activité
Calme
Clarté
Clairement spécifiée
Accessibilité débutants
68/100

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.