developing-cast

name	developing-cast
description	Implements LangGraph cast components following systematic workflow (state, deps, nodes, conditions, graph). Use when implementing cast, building nodes/agents/tools, need LangGraph patterns (memory, retry, guardrails, vector stores, node timeouts, error handlers, DeltaChannel, graceful shutdown), or ask "implement cast", "build graph", "add node".
version	2026.05.27
author	Proact0
allowed-tools	["Bash(uv sync *)","Bash(uv add --dev *)","Bash(uv add --group *)","Bash(uv add --package *)","Bash(uv remove --package *)","Read","Write","Edit","AskUserQuestion"]

Developing Backend's Cast

Implement LangGraph casts following Backend Act patterns.

When to Use

• Building nodes, agents, tools, or graphs
• Need LangGraph implementation patterns
• Have architecture specs to implement

When NOT to Use

• Architecture design → architecting-act
• DeepAgent harness (create_deep_agent, backends, subagents) → developing-deepagent
• Project / cast scaffolding → run uv run act new (project) or uv run act cast (new cast) directly
• Testing → testing-cast

---

Implementation Workflow

Step 1: Understand CLAUDE.md

Act = Project, Cast = Graph/Workflow in the project

If CLAUDE.md exists:

1. Read /CLAUDE.md → Act overview, find target cast
2. Read /casts/{cast_slug}/CLAUDE.md → Architecture diagram, node specs
3. Proceed to Step 2

If CLAUDE.md not found: AskUserQuestion Format:

{
  "question": "CLAUDE.md not found. Create architecture first?",
  "header": "Architecture",
  "multiSelect": false,
  "options": [
    {"label": "Yes", "description": "Switch to architecting-act skill"},
    {"label": "No", "description": "Proceed without architecture specs"}
  ]
}

• Yes → use architecting-act skill first
• No → proceed to Step 2

Step 2: Implementation

Implement in order: state → dependency modules → nodes → conditions → graph

• Use Component Reference tables below for each component type

1. State (state.py)           # Foundation
   ↓
2. Dependency modules         # agents, models, tools, prompts, middlewares, utils (if needed)
   ↓
3. Nodes (nodes.py)           # Business logic
   ↓
4. Conditions (conditions.py) # Route Functions (if needed)
   ↓
5. Graph (graph.py)           # Assembly

Optional Step 3: Create required environment variables (if needed)

Update .env.example (project root):

OPENAI_API_KEY=your_key
ANTHROPIC_API_KEY=your_key

Optional Step 4: Install dependency packages (if needed)

The uv workspace exposes each cast under casts/* as a member; install per-cast dependencies into the cast's own pyproject.toml:

# Cast-scoped dependency (most common — the dep is used by one cast)
uv add --package {{ cookiecutter.cast_slug }} <package-name>

# Project-scoped dependency (rare — shared across all casts)
uv add --package {{ cookiecutter.act_slug }} <package-name>

---

Component Reference

Core Components

Use when	Resource
defining graph state with TypedDict	core/state.md
implementing sync/async node classes	core/node.md
using prebuilt ToolNode for tool execution	core/tool-node.md
setting up edges or conditional routing	core/edge.md
assembling StateGraph and compiling	core/graph.md
reusing graphs as subgraphs	core/subgraph.md

Prompts & Messages

Use when	Resource
creating System/Human/AI/Tool messages	prompts/message-types.md
handling image/audio/PDF inputs	prompts/multimodal.md

Models & Agents

Use when	Resource
choosing between OpenAI/Anthropic/Google	models/select-chat-models.md
configuring model (temperature, tokens)	models/standalone-model.md
need model to return structured output(Pydantic schema)	models/structured-output.md
creating agent with tools	agents/configuration.md
need agent to return structured output(Pydantic schema)	agents/structured-output.md

Tools

Use when	Resource
creating simple tool with @tool	tools/basic-tool.md
tool needs complex Pydantic inputs	tools/tool-with-complex-inputs.md
tool needs to read/write state or store	tools/access-context.md

Memory

Use when	Resource
adding conversation memory to agent	memory/short-term/add-to-agent.md
customizing agent memory storage	memory/short-term/customize-agent-memory.md
trimming/deleting/summarizing history	memory/short-term/manage-conversations.md
accessing memory from middleware/tools	memory/short-term/access-and-modify-memory.md
persisting data across sessions (Store)	memory/long-term/memory-storage.md
accessing Store from within tools	memory/long-term/in-tools.md

Middleware - Reliability

Use when	Resource
LLM calls fail intermittently	middlewares/provider-agnostic/model-retry.md
tool execution fails intermittently	middlewares/provider-agnostic/tool-retry.md
need backup model when primary fails	middlewares/provider-agnostic/model-fallback.md

Middleware - Safety & Control

Use when	Resource
validating/blocking inappropriate content	middlewares/provider-agnostic/guardrails.md
preventing infinite LLM call loops	middlewares/provider-agnostic/model-call-limit.md
limiting tool calls to control costs	middlewares/provider-agnostic/tool-call-limit.md
requiring human approval at checkpoints	middlewares/provider-agnostic/human-in-the-loop.md

Middleware - Tool Management

Use when	Resource
dynamically selecting relevant tools	middlewares/provider-agnostic/llm-tool-selector.md
emulating tools with LLM for testing	middlewares/provider-agnostic/llm-tool-emulator.md
agent needs persistent shell session	middlewares/provider-agnostic/shell-tool.md
agent needs to search files (glob/grep)	middlewares/provider-agnostic/file-search.md
agent needs task planning/tracking	middlewares/provider-agnostic/to-do-list.md

Middleware - Context

Use when	Resource
modifying/removing messages at runtime	middlewares/provider-agnostic/context-editing.md
auto-summarizing near token limits	middlewares/provider-agnostic/summarization.md

Middleware - Provider-Specific

Use when	Resource
using OpenAI moderation API	middlewares/provider-specific/openai.md
using Claude caching/bash/text-editor	middlewares/provider-specific/anthropic.md
building custom before/after/wrap hooks	middlewares/custom.md

Integrations

Use when	Resource
converting text to embedding vectors	integrations/embedding.md
using FAISS/Pinecone/Chroma stores	integrations/vector-stores.md
splitting long documents into chunks	integrations/text-splitter.md

---

Verification

• CLAUDE.md checked (root + cast-specific if exists, skipped if not)
• Implementation order: state → deps(option) → nodes → conditions(option) → graph
• Nodes match CLAUDE.md specs
• Dependencies installed, env vars configured
• Graph compiles