Strands Coder

Autonomous GitHub Agent for Strands Agents

GitHub Marketplace ◆ Strands Docs ◆ AWS Bedrock ◆ Langfuse ◆ Strands SDK ◆ Dashboard

An autonomous GitHub agent that codes, reviews, tracks, schedules, and evolves—all on its own. Built with Strands Agents SDK, it doesn't just respond to events—it proactively identifies work, creates issues, reviews PRs, tracks progress in GitHub Projects, and learns from every interaction.

🚀 Get Started in 2 Minutes

Available on GitHub Marketplace

1. Add the workflow to your repo

Create .github/workflows/agent.yml:

name: AI Agent

on:
  issues:
    types: [opened, edited]
  issue_comment:
    types: [created]
  pull_request:
    types: [opened, synchronize]
  workflow_dispatch:
    inputs:
      prompt:
        description: 'Task for the agent'
        required: true

jobs:
  agent:
    runs-on: ubuntu-latest
    permissions:
      contents: write
      issues: write
      pull-requests: write
      id-token: write
    steps:
      - uses: cagataycali/strands-coder@v1
        with:
          prompt: ${{ github.event.inputs.prompt || 'Respond to this GitHub event' }}
          provider: bedrock
          aws_role_arn: ${{ secrets.AWS_ROLE_ARN }}
        env:
          PAT_TOKEN: ${{ secrets.PAT_TOKEN }}

2. Add secrets

Secret	Description
`AWS_ROLE_ARN`	ARN of IAM role with Bedrock access (setup guide)
`PAT_TOKEN`	GitHub Personal Access Token with repo permissions

3. Trigger the agent

Open an issue → Agent responds automatically
Comment on PR → Agent reviews and suggests
Manual dispatch → Run any prompt on-demand

💡 Alternative providers: Set provider: anthropic with ANTHROPIC_API_KEY, or provider: openai with OPENAI_API_KEY

How It Works

graph LR
    A[GitHub Event<br/>'Issue opened'] --> B[Strands Agent]
    B --> C[Context Injection<br/>GraphQL]
    C --> D[Knowledge Base<br/>RAG Retrieval]
    D --> E[Agent Reasoning<br/>Claude]
    E --> F[Tool Execution<br/>GitHub/Projects]
    F --> G[Knowledge Storage<br/>Learning]
    G --> H[Self-Evolution<br/>Prompt Updates]

    classDef input fill:#2ea44f,stroke:#1b7735,color:#fff
    classDef agent fill:#0969da,stroke:#044289,color:#fff
    classDef knowledge fill:#8250df,stroke:#5a32a3,color:#fff
    classDef action fill:#bf8700,stroke:#875e00,color:#fff

    class A input
    class B,E agent
    class C,D,G knowledge
    class F,H action

Architecture

flowchart TB
    subgraph Triggers["⚡ Event Triggers"]
        GH[GitHub Events]
        SC[Schedule: */4 hours]
        CL[Control Loop: Hourly]
        WD[Workflow Dispatch]
    end

    subgraph Agent["🤖 Strands Agent"]
        direction TB
        AR[Agent Runner]
        CTX[Context Builder]
        SP[System Prompt]
    end

    subgraph Tools["🔧 Tool Registry"]
        direction TB
        UG[use_github]
        PR[projects]
        SCH[scheduler]
        SUB[create_subagent]
        KB[store_in_kb]
        SYS[system_prompt]
    end

    subgraph External["☁️ External Services"]
        direction TB
        GQL[GitHub GraphQL]
        BKB[Bedrock KB]
        LF[Langfuse]
        S3[S3 Sessions]
    end

    GH --> AR
    SC --> AR
    CL --> WD
    WD --> AR
    AR --> CTX
    CTX --> SP
    AR --> Tools
    UG --> GQL
    PR --> GQL
    KB --> BKB
    AR --> LF
    AR --> S3

    classDef triggerStyle fill:#2ea44f,stroke:#1b7735,color:#fff
    classDef agentStyle fill:#0969da,stroke:#044289,color:#fff
    classDef toolStyle fill:#8250df,stroke:#5a32a3,color:#fff
    classDef externalStyle fill:#bf8700,stroke:#875e00,color:#fff

    class GH,SC,CL,WD triggerStyle
    class AR,CTX,SP agentStyle
    class UG,PR,SCH,SUB,KB,SYS toolStyle
    class GQL,BKB,LF,S3 externalStyle

🌐 Web Dashboard

Live Demo: dev.strands.my

A zero-install PWA dashboard for managing your Strands Coder agent:

Tab	Features
📊 Dashboard	Project stats, GitHub Project board integration
📋 Issues	Browse, filter, create issues with full thread view
⚡ Actions	Live log streaming, issue-linked workflow runs
📈 Traces	Langfuse integration with conversation playback
🤖 Agent	Configure model, tools, MCP servers, knowledge base
📅 Schedule	Cron job management with enable/disable
⚙️ Settings	GitHub & Langfuse config, encrypted share links

Dashboard Setup

Open dev.strands.my on any device
Configure GitHub token and repository in Settings
Optionally add Langfuse credentials for trace viewing
Share encrypted config links with your team

The dashboard connects directly to GitHub and Langfuse APIs from your browser—no backend required.

Self-Host: Deploy docs/index.html to any static hosting (GitHub Pages, Cloudflare, etc.)

Quick Start

from strands import Agent
from strands_coder import use_github, projects, scheduler, store_in_kb

# Create agent with GitHub tools
agent = Agent(
    tools=[use_github, projects, scheduler, store_in_kb],
    system_prompt="You are an autonomous GitHub agent."
)

# Review open PRs
agent("List all open pull requests and summarize their status")

# Add issue to project board
agent("Add issue #42 to the project and set status to 'In Progress'")

# Schedule recurring task
agent("Schedule a daily code review job at 9 AM UTC")

Installation

pip install strands-coder

Or from source:

git clone https://github.com/cagataycali/strands-coder
cd strands-coder
pip install -e .

🔐 GitHub Actions Setup (Recommended)

1. Configure Secrets

# Repository Settings → Secrets and variables → Actions → Secrets
AWS_ROLE_ARN: arn:aws:iam::YOUR_ACCOUNT:role/StrandsCoder
PAT_TOKEN: ghp_xxxxxxxxxxxxx
AUTHORIZED_USERS: octocat,yourname,github-actions[bot]

# Optional - Observability
LANGFUSE_BASE_URL: https://cloud.langfuse.com
LANGFUSE_PUBLIC_KEY: pk-lf-xxx
LANGFUSE_SECRET_KEY: sk-lf-xxx

2. Configure Variables

# Repository Settings → Secrets and variables → Actions → Variables
SYSTEM_PROMPT: "Your custom instructions here"
STRANDS_CODER_PROJECT_ID: PVT_kwDOAbcDefg
STRANDS_KNOWLEDGE_BASE_ID: YOUR_KB_ID
STRANDS_MODEL_ID: global.anthropic.claude-opus-4-5-20251101-v1:0
STRANDS_TOOLS: strands_tools:shell,retrieve;strands_coder:use_github,projects,scheduler

3. Deploy Workflows

git add .github/workflows/agent.yml .github/workflows/control.yml
git commit -m "feat: add strands-coder agent"
git push origin main

Event Flow

sequenceDiagram
    participant User
    participant GitHub as GitHub Event
    participant Agent as Strands Agent
    participant KB as Knowledge Base
    participant Tools as Tool Registry

    User->>GitHub: Opens Issue #42
    GitHub->>Agent: Trigger workflow
    
    Agent->>Agent: fetch_github_event_context()
    Note over Agent: GraphQL: Full thread + comments
    
    Agent->>KB: retrieve(issue_content)
    KB-->>Agent: Similar issues & solutions
    
    Agent->>Agent: Reasoning with context
    
    Agent->>Tools: use_github(add_comment)
    Agent->>Tools: projects(update_status)
    
    Agent->>KB: store_in_kb(conversation)
    Note over KB: Learning for future
    
    Agent-->>GitHub: Comment posted
    GitHub-->>User: Notification

Tools Reference

use_github

Universal GitHub GraphQL API interface for queries and mutations.

Action	Parameters	Description	Example
`query`	`query`, `variables`	Execute GraphQL query	`"Get repository info"`
`mutation`	`query`, `variables`	Execute GraphQL mutation	`"Create issue"`

Natural Language Examples:

# Query repository
agent("List all open issues in this repository")

# Create PR comment
agent("Add a review comment to PR #45 suggesting code improvements")

# Close issue
agent("Close issue #42 with a summary comment")

GraphQL Query Examples

# Get repository info
agent.tool.use_github(
    query_type="query",
    query="""
    query($owner: String!, $name: String!) {
      repository(owner: $owner, name: $name) {
        name
        description
        stargazerCount
        issues(states: OPEN) { totalCount }
      }
    }
    """,
    variables={"owner": "strands-agents", "name": "sdk-python"},
    label="Get repository info"
)

projects

GitHub Projects V2 management for cross-repository work tracking.

Action	Parameters	Description	Example
`list_projects`	`owner`	List all projects	`"List my projects"`
`get_project`	`project_id`	Get project details	`"Show project status"`
`add_issue`	`repository`, `issue_number`	Add issue to project	`"Track issue #42"`
`add_pr`	`repository`, `pr_number`	Add PR to project	`"Track PR #45"`
`update_item`	`item_id`, `field_name`, `field_value`	Update item field	`"Set status to Done"`
`get_progress`	-	Get progress summary	`"Show project progress"`

Natural Language Examples:

# Add issue to project
agent("Add issue #42 from strands-agents/sdk-python to my project")

# Update status
agent("Mark the item for issue #42 as 'In Progress'")

# Get summary
agent("Show me the project progress with counts by status")

scheduler

Cron-based and one-time job scheduling via GitHub Actions variables.

Action	Parameters	Description	Example
`list`	-	List all scheduled jobs	`"List scheduled jobs"`
`add`	`job_id`, `cron`/`run_at`, `prompt`	Add new job	`"Schedule daily review"`
`remove`	`job_id`	Remove job	`"Remove daily_review job"`
`enable`	`job_id`	Enable disabled job	`"Enable weekly_report"`
`disable`	`job_id`	Disable job	`"Pause daily_review"`
`check`	-	Check jobs due now	`"What jobs should run?"`

Natural Language Examples:

# Schedule recurring job
agent("Schedule a job called 'daily_review' to run at 9 AM UTC daily with prompt 'Review open PRs'")

# One-time scheduled job
agent("Schedule deployment for January 20th at 2 PM UTC, auto-remove after execution")

# Disable temporarily
agent("Disable the daily_review job")

AGENT_SCHEDULES Format

{
  "jobs": {
    "daily_review": {
      "cron": "0 9 * * *",
      "enabled": true,
      "prompt": "Review open PRs and provide feedback",
      "system_prompt": "You are a code reviewer",
      "tools": "strands_tools:shell;strands_coder:use_github",
      "model": "global.anthropic.claude-sonnet-4-5-20251101-v1:0"
    },
    "friday_deploy": {
      "run_at": "2026-01-17T15:00:00Z",
      "prompt": "Deploy release to production",
      "once": true
    }
  },
  "timezone": "UTC"
}

Cron Expression Reference:

┌───────────── minute (0-59)
│ ┌─────────── hour (0-23)
│ │ ┌───────── day of month (1-31)
│ │ │ ┌─────── month (1-12)
│ │ │ │ ┌───── day of week (0-6, Sunday=0)
│ │ │ │ │
* * * * *

0 * * * *     # Every hour
0 9 * * *     # Daily at 9:00 AM
0 9 * * 1-5   # Weekdays at 9:00 AM
0 */4 * * *   # Every 4 hours

create_subagent

Spawn parallel sub-agents via GitHub Actions workflow dispatch.

Action	Parameters	Description	Example
`create`	`repository`, `workflow_id`, `prompt`, `model`, `tools`	Spawn sub-agent	`"Create review agent"`
`status`	`repository`, `run_id`	Check run status	`"Check agent status"`
`list`	`repository`, `workflow_id`	List recent runs	`"List sub-agent runs"`

Natural Language Examples:

# Spawn specialized sub-agent
agent("Create a sub-agent to review PR #42 for security issues using Claude Opus")

# Check status
agent("Check the status of sub-agent run 12345678")

Sub-Agent Configuration

agent.tool.create_subagent(
    repository="owner/repo",
    workflow_id="agent.yml",
    prompt="Review PR #42 for security vulnerabilities",
    model="global.anthropic.claude-opus-4-5-20251101-v1:0",
    max_tokens=60000,
    tools="strands_tools:shell,retrieve",
    system_prompt="You are a security expert focused on code review"
)

store_in_kb

Async knowledge base storage for persistent memory across sessions.

Action	Parameters	Description	Example
`store`	`content`, `title`	Store content	`"Remember this solution"`

Natural Language Examples:

# Store learning
agent("Store in knowledge base: 'Memory leaks fixed by implementing proper cleanup in __del__'")

system_prompt

Self-modification tool for runtime prompt updates.

Action	Parameters	Description	Example
`view`	-	View current prompt	`"Show system prompt"`
`update`	`prompt`, `repository`	Replace prompt	`"Update my instructions"`
`add_context`	`context`, `repository`	Append context	`"Remember this pattern"`
`reset`	-	Reset to default	`"Reset system prompt"`

Natural Language Examples:

# Add learning
agent("Add to system prompt: 'Users prefer detailed code explanations with examples'")

# Sync to GitHub
agent("Update system prompt and sync to repository variable")

Context Injection

The agent sees full conversation context, not just the latest message:

flowchart LR
    subgraph Event["📥 GitHub Event"]
        E1[Issue Created]
        E2[Comment Added]
        E3[PR Reviewed]
    end

    subgraph Context["📋 Context Builder"]
        direction TB
        C1[Raw GITHUB_CONTEXT]
        C2[GraphQL Enrichment]
        C3[Comments Thread]
        C4[Linked Items]
        C5[Project Status]
    end

    subgraph Prompt["🧠 System Prompt"]
        P1[Base Instructions]
        P2[Event Context]
        P3[Project Context]
        P4[Self-Awareness]
    end

    Event --> Context
    Context --> Prompt
    Prompt --> Agent[Agent]

    classDef eventStyle fill:#2ea44f,stroke:#1b7735,color:#fff
    classDef contextStyle fill:#8250df,stroke:#5a32a3,color:#fff
    classDef promptStyle fill:#0969da,stroke:#044289,color:#fff

    class E1,E2,E3 eventStyle
    class C1,C2,C3,C4,C5 contextStyle
    class P1,P2,P3,P4 promptStyle

Example Injected Context

## 📋 RAW GITHUB_CONTEXT
**Event Type:** `issue_comment`
**Repository:** `owner/repo`
**Action:** `created`
**Actor:** `alice`

## 🎫 ISSUE CONTEXT
**Issue:** #42: Memory leak in agent loop
**State:** open | **Author:** @alice | **Created:** 2026-01-09

### Original Issue Body
The agent seems to leak memory when running for extended periods...

### 💬 Comments (5 total)
**Comment #1** by @bob at 2026-01-09 14:32:
I've seen this too. Here's a stack trace...

**Comment #2** by @strands-agent at 2026-01-09 15:15:
I've analyzed the code and found the issue in the event loop...

### 🔗 Linked Items
- PR #45: Fix memory leak (OPEN)

## 📊 PROJECT CONTEXT
**Project:** Strands-Coder (#1)
- Todo: 5 | In Progress: 3 | Done: 42

Two-Workflow Architecture

flowchart TB
    subgraph agent["agent.yml (Main Agent)"]
        A1[GitHub Events] --> A2[Event Handler]
        A3[Schedule: */4 hours] --> A2
        A4[Workflow Dispatch] --> A2
        A2 --> A5[Agent Runner]
        A5 --> A6[Tool Execution]
    end

    subgraph control["control.yml (Scheduler)"]
        C1[Schedule: Hourly] --> C2[Parse AGENT_SCHEDULES]
        C2 --> C3{Jobs Due?}
        C3 -->|Yes| C4[Dispatch agent.yml]
        C3 -->|No| C5[Skip]
    end

    C4 --> A4

    classDef mainStyle fill:#0969da,stroke:#044289,color:#fff
    classDef controlStyle fill:#8250df,stroke:#5a32a3,color:#fff

    class A1,A2,A3,A4,A5,A6 mainStyle
    class C1,C2,C3,C4,C5 controlStyle

Workflow	Trigger	Purpose
`agent.yml`	Events, Schedule, Dispatch	Main agent execution
`control.yml`	Hourly	Check `AGENT_SCHEDULES` and dispatch jobs

Observability

flowchart LR
    subgraph Agent["🤖 Agent Execution"]
        A1[Start] --> A2[Reasoning]
        A2 --> A3[Tool Calls]
        A3 --> A4[Complete]
    end

    subgraph Telemetry["📊 Telemetry"]
        T1[OTEL Exporter]
        T2[Trace Attributes]
        T3[Issue Linking]
    end

    subgraph Langfuse["🔭 Langfuse"]
        L1[Traces]
        L2[Generations]
        L3[Tool Calls]
        L4[Costs]
    end

    Agent --> Telemetry
    Telemetry --> Langfuse

    classDef agentStyle fill:#2ea44f,stroke:#1b7735,color:#fff
    classDef telStyle fill:#bf8700,stroke:#875e00,color:#fff
    classDef lfStyle fill:#4B32C3,stroke:#3a2799,color:#fff

    class A1,A2,A3,A4 agentStyle
    class T1,T2,T3 telStyle
    class L1,L2,L3,L4 lfStyle

Features:

🏷️ Trace Linking: All traces tagged with issue:{number} for filtering
💬 Conversation View: Timeline of agent reasoning
💰 Cost Tracking: Token usage and costs per trace
🔄 Session Persistence: S3-backed conversation history

Dashboard Features

The web dashboard at dev.strands.my (also available as docs/index.html) provides:

Tab	Description
📋 Issues	Browse, search, create issues
📊 Dashboard	Project stats and board overview
⚡ Actions	Workflow runs with live logs
📈 Traces	Langfuse traces with playback
📅 Schedule	Job scheduler management
🤖 Agent	Model, tools, prompt configuration
⚙️ Settings	GitHub & Langfuse credentials

Project Structure

strands-coder/
├── strands_coder/
│   ├── __init__.py              # Package exports
│   ├── agent_runner.py          # Main entry point
│   ├── context.py               # Context builder & prompt assembly
│   └── tools/
│       ├── __init__.py          # Tool exports
│       ├── use_github.py        # GitHub GraphQL interface
│       ├── github_tools.py      # GitHub REST helpers
│       ├── projects.py          # GitHub Projects V2
│       ├── scheduler.py         # Job scheduling
│       ├── create_subagent.py   # Parallel agent spawning
│       ├── store_in_kb.py       # Knowledge base storage
│       └── system_prompt.py     # Self-modification
├── .github/
│   └── workflows/
│       ├── agent.yml            # Main agent workflow
│       └── control.yml          # Scheduler control loop
├── docs/
│   └── index.html               # Web dashboard
├── action.yml                   # GitHub Action definition
└── pyproject.toml               # Package configuration

Model Configuration

Model	Use Case	Max Tokens
`global.anthropic.claude-opus-4-5-20251101-v1:0`	Complex reasoning	60,000
`global.anthropic.claude-sonnet-4-5-20251101-v1:0`	Balanced performance	60,000
`us.anthropic.claude-sonnet-4-20250514-v1:0`	Fast & efficient	32,768

MCP Server Configuration

MCP_SERVERS: |
  {
    "mcpServers": {
      "strands-docs": {
        "command": "uvx",
        "args": ["strands-agents-mcp-server"],
        "disabled": false,
        "disabledTools": []
      },
      "custom-server": {
        "url": "http://localhost:8000/mcp"
      }
    }
  }

Example: Complete Workflow

#!/usr/bin/env python3
"""Autonomous GitHub agent example."""
from strands import Agent
from strands_coder import (
    use_github,
    projects,
    scheduler,
    create_subagent,
    store_in_kb,
    system_prompt,
)

# Create agent with full toolset
agent = Agent(
    tools=[use_github, projects, scheduler, create_subagent, store_in_kb, system_prompt],
    system_prompt="""You are an autonomous GitHub agent.
    - Review PRs with detailed feedback
    - Track work in GitHub Projects
    - Learn from every interaction
    - Schedule recurring maintenance tasks
    """
)

# Interactive mode
while True:
    user_input = input("\n🤖 > ")
    if user_input.lower() in ["exit", "quit"]:
        break
    agent(user_input)

CLI Usage

# Run with prompt
strands-coder "Review all open PRs"

# Via environment variable
STRANDS_PROMPT="Create tracking issues for stale PRs" strands-coder

# With custom configuration
STRANDS_MODEL_ID="global.anthropic.claude-opus-4-5-20251101-v1:0" \
STRANDS_TOOLS="strands_coder:use_github,projects" \
strands-coder "Analyze repository health"

Contributing

We welcome contributions! The agent will review your PR automatically. 🧬

# 1. Fork & clone
git clone https://github.com/YOUR_USERNAME/strands-coder.git

# 2. Install dev dependencies
pip install -e ".[dev]"

# 3. Run tests
hatch run test

# 4. Format & lint
hatch run prepare

# 5. Submit PR
git push origin feat/my-feature

License

Apache-2.0 - see LICENSE file.

Links

Marketplace ◆ GitHub ◆ PyPI ◆ Dashboard ◆ Strands Docs ◆ AWS Bedrock ◆ Langfuse

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
.github/workflows		.github/workflows
docs		docs
strands_coder		strands_coder
tools		tools
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
SYSTEM_PROMPT.md		SYSTEM_PROMPT.md
action.yml		action.yml
pyproject.toml		pyproject.toml
setup-aws-oidc.sh		setup-aws-oidc.sh

Folders and files

Latest commit

History

Repository files navigation

Strands Coder

Autonomous GitHub Agent for Strands Agents

🚀 Get Started in 2 Minutes

1. Add the workflow to your repo

2. Add secrets

3. Trigger the agent

How It Works

Architecture

🌐 Web Dashboard

Quick Start

Installation

1. Configure Secrets

2. Configure Variables

3. Deploy Workflows

Event Flow

Tools Reference

use_github

projects

scheduler

create_subagent

store_in_kb

system_prompt

Context Injection

Two-Workflow Architecture

Observability

Project Structure

Model Configuration

Example: Complete Workflow

CLI Usage

Contributing

License

Links

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Contributors

Uh oh!

Languages