orchestrating-agents

What this skill does

## SURFACE ROUTING — read first

This skill hand-rolls subagent orchestration via raw Anthropic API calls. A
managed runtime now does the same job. Which one to use depends on your surface:

- **In Claude Code (incl. CCotw): use the native runtime, NOT this skill.** If you
  can invoke `/deep-research`, trigger a run with the `workflow` keyword, set
  `/effort ultracode`, or spawn Task subagents — do that instead. The runtime gives
  16-concurrent / 1000-agent ceilings, an approval gate, adversarial cross-review,
  and in-session resume that this skill would otherwise reimplement badly. Dynamic
  workflows shipped in research preview (Claude Code v2.1.154+, 2026).
- **In claude.ai chat or the bare API (no workflow runtime): use this skill.**
  Parallel API instances over httpx is the only fan-out path here. Proceed below.

Discriminator: do you have a native subagent/Task tool or a workflow command? Yes
→ native. No → this skill. Never reimplement the runtime where it already exists.

# Orchestrating Agents

This skill enables programmatic API invocations for advanced workflows including parallel processing, task delegation, and multi-agent analysis using the Anthropic API.

## When to Use This Skill

**Primary use cases:**
- **Parallel sub-tasks**: Break complex analysis into simultaneous independent streams
- **Multi-perspective analysis**: Get 3-5 different expert viewpoints concurrently
- **Delegation**: Offload specific subtasks to specialized API instances
- **Recursive workflows**: Orchestrator coordinating multiple API instances
- **High-volume processing**: Batch process multiple items concurrently

**Trigger patterns:**
- "Parallel analysis", "multi-perspective review", "concurrent processing"
- "Delegate subtasks", "coordinate multiple agents"
- "Run analyses from different perspectives"
- "Get expert opinions from multiple angles"

## Quick Start

### Single Invocation

```python
import sys
sys.path.append('/home/user/claude-skills/orchestrating-agents/scripts')
from claude_client import invoke_claude

response = invoke_claude(
    prompt="Analyze this code for security vulnerabilities: ...",
    model="claude-sonnet-4-6"
)
print(response)
```

### Parallel Multi-Perspective Analysis

```python
from claude_client import invoke_parallel

prompts = [
    {
        "prompt": "Analyze from security perspective: ...",
        "system": "You are a security expert"
    },
    {
        "prompt": "Analyze from performance perspective: ...",
        "system": "You are a performance optimization expert"
    },
    {
        "prompt": "Analyze from maintainability perspective: ...",
        "system": "You are a software architecture expert"
    }
]

results = invoke_parallel(prompts, model="claude-sonnet-4-6")

for i, result in enumerate(results):
    print(f"\n=== Perspective {i+1} ===")
    print(result)
```

### Parallel with Shared Cached Context (Recommended)

For parallel operations with shared base context, use caching to reduce costs by up to 90%:

```python
from claude_client import invoke_parallel

# Large context shared across all sub-agents (e.g., codebase, documentation)
base_context = """
<codebase>
...large codebase or documentation (1000+ tokens)...
</codebase>
"""

prompts = [
    {"prompt": "Find security vulnerabilities in the authentication module"},
    {"prompt": "Identify performance bottlenecks in the API layer"},
    {"prompt": "Suggest refactoring opportunities in the database layer"}
]

# First sub-agent creates cache, subsequent ones reuse it
results = invoke_parallel(
    prompts,
    shared_system=base_context,
    cache_shared_system=True  # 90% cost reduction for cached content
)
```

### Multi-Turn Conversation with Auto-Caching

For sub-agents that need multiple rounds of conversation:

```python
from claude_client import ConversationThread

# Create a conversation thread (auto-caches history)
agent = ConversationThread(
    system="You are a code refactoring expert with access to the codebase",
    cache_system=True
)

# Turn 1: Initial analysis
response1 = agent.send("Analyze the UserAuth class for issues")
print(response1)

# Turn 2: Follow-up (reuses cached system + turn 1)
response2 = agent.send("How would you refactor the login method?")
print(response2)

# Turn 3: Implementation (reuses all previous context)
response3 = agent.send("Show me the refactored code")
print(response3)
```

### Streaming Responses

For real-time feedback from sub-agents:

```python
from claude_client import invoke_claude_streaming

def show_progress(chunk):
    print(chunk, end='', flush=True)

response = invoke_claude_streaming(
    "Write a comprehensive security analysis...",
    callback=show_progress
)
```

### Parallel Streaming

Monitor multiple sub-agents simultaneously:

```python
from claude_client import invoke_parallel_streaming

def agent1_callback(chunk):
    print(f"[Security] {chunk}", end='', flush=True)

def agent2_callback(chunk):
    print(f"[Performance] {chunk}", end='', flush=True)

results = invoke_parallel_streaming(
    [
        {"prompt": "Security review: ..."},
        {"prompt": "Performance review: ..."}
    ],
    callbacks=[agent1_callback, agent2_callback]
)
```

### Interruptible Operations

Cancel long-running parallel operations:

```python
from claude_client import invoke_parallel_interruptible, InterruptToken
import threading
import time

token = InterruptToken()

# Run in background
def run_analysis():
    results = invoke_parallel_interruptible(
        prompts=[...],
        interrupt_token=token
    )
    return results

thread = threading.Thread(target=run_analysis)
thread.start()

# Interrupt after 5 seconds
time.sleep(5)
token.interrupt()
```

## Core Functions

### `invoke_claude()`

Single synchronous invocation with full control:

```python
invoke_claude(
    prompt: str | list[dict],
    model: str = "claude-sonnet-4-6",
    system: str | list[dict] | None = None,
    max_tokens: int = 4096,
    temperature: float = 1.0,
    streaming: bool = False,
    cache_system: bool = False,
    cache_prompt: bool = False,
    messages: list[dict] | None = None,
    **kwargs
) -> str
```

**Parameters:**
- `prompt`: The user message (string or list of content blocks)
- `model`: Claude model to use (default: claude-sonnet-4-6)
- `system`: Optional system prompt (string or list of content blocks)
- `max_tokens`: Maximum tokens in response (default: 4096)
- `temperature`: Randomness 0-1 (default: 1.0)
- `streaming`: Enable streaming response (default: False)
- `cache_system`: Add cache_control to system prompt (requires 1024+ tokens, default: False)
- `cache_prompt`: Add cache_control to user prompt (requires 1024+ tokens, default: False)
- `messages`: Pre-built messages list for multi-turn (overrides prompt)
- `**kwargs`: Additional API parameters (top_p, top_k, etc.)

**Returns:** Response text as string

**Note:** Caching requires minimum 1,024 tokens per cache breakpoint. Cache lifetime is 5 minutes (refreshed on use).

### `invoke_parallel()`

Concurrent invocations using lightweight workflow pattern:

```python
invoke_parallel(
    prompts: list[dict],
    model: str = "claude-sonnet-4-6",
    max_tokens: int = 4096,
    max_workers: int = 5,
    shared_system: str | list[dict] | None = None,
    cache_shared_system: bool = False
) -> list[str]
```

**Parameters:**
- `prompts`: List of dicts with 'prompt' (required) and optional 'system', 'temperature', 'cache_system', 'cache_prompt', etc.
- `model`: Claude model for all invocations
- `max_tokens`: Max tokens per response
- `max_workers`: Max concurrent API calls (default: 5, max: 10)
- `shared_system`: System context shared across ALL invocations (for cache efficiency)
- `cache_shared_system`: Add cache_control to shared_system (default: False)

**Returns:** List of response strings in same order as prompts

**Note:** For optimal cost savings, put large common context (1024+ tokens) in `shared_system` with `cache_shared_system=True`. Fir