Play 51 — Autonomous Coding Agent

Autonomous issue-to-PR pipeline — analyzes GitHub issues, indexes codebase, creates implementation plans, generates multi-file code changes, auto-generates tests, self-heals failing CI, and creates PRs with full descriptions. Human-in-the-loop approval gates with optional auto-merge for small bug fixes.

Architecture

Component	Technology	Purpose
Code Generation	Azure OpenAI (GPT-4o)	Implementation, test generation, self-healing
Codebase Index	Custom (imports/exports/structure)	Understand repo before coding
GitHub Integration	GitHub API + webhooks	Issue reading, branch/PR creation
Agent Runtime	Azure Container Apps	Pipeline orchestration
CI Integration	GitHub Actions	Test execution, lint verification
Secrets	Azure Key Vault	GitHub PAT, OpenAI key

📐 Full architecture details

Aspect	Play 24 (Code Review)	Play 51 (Autonomous Coding)	Play 37 (AI DevOps)
Direction	Reviews existing code	Generates new code from issues	Incident response
Input	Pull request diff	GitHub issue description	Alert/incident
Output	Review comments	Complete PR (code + tests + description)	Runbook execution
Trigger	PR created	Issue labeled `auto-fix`	Alert fired
Scope	Read-only analysis	Write: creates branches, commits, PRs	Remediation commands
Testing	Reviews test quality	Generates tests for changed code	Verifies resolution

DevKit Structure


51-autonomous-coding-agent/
├── agent.md                                  # Root orchestrator with handoffs
├── .github/
│   ├── copilot-instructions.md               # Domain knowledge (<150 lines)
│   ├── agents/
│   │   ├── builder.agent.md                  # Issue→plan→code→test→PR pipeline
│   │   ├── reviewer.agent.md                 # Code quality, test coverage, scope
│   │   └── tuner.agent.md                    # Plan accuracy, iterations, cost
│   ├── prompts/
│   │   ├── deploy.prompt.md                  # Deploy agent + GitHub webhook
│   │   ├── test.prompt.md                    # Resolve sample issue end-to-end
│   │   ├── review.prompt.md                  # Audit code quality + scope
│   │   └── evaluate.prompt.md               # Measure resolution rate + cost
│   ├── skills/
│   │   ├── deploy-autonomous-coding-agent/   # Pipeline + codebase index + webhook
│   │   ├── evaluate-autonomous-coding-agent/ # Resolution, quality, tests, PRs, cost
│   │   └── tune-autonomous-coding-agent/     # Model per task, scope, auto-merge
│   └── instructions/
│       └── autonomous-coding-agent-patterns.instructions.md
├── config/                                   # TuneKit
│   ├── openai.json                           # Model per task (code=gpt-4o, tests=mini)
│   ├── guardrails.json                       # Max files, scope, quality gates
│   └── agents.json                           # Auto-merge, indexing, branch naming
├── infra/                                    # Bicep IaC
│   ├── main.bicep
│   └── parameters.json
└── spec/                                     # SpecKit
    └── fai-manifest.json

Quick Start


# 1. Deploy agent + configure GitHub webhook
/deploy
 
# 2. Resolve a sample issue end-to-end
/test
 
# 3. Audit code quality and scope
/review
 
# 4. Measure resolution rate and cost
/evaluate

Key Metrics

Metric	Target	Description
Resolution Rate	> 70%	Issues fully resolved to merged PR
Code Compiles	> 95%	Generated code passes linter
Test Generated	> 90%	Tests created for changed code
PR Acceptance	> 60%	PRs approved by human reviewer
Avg Iterations	< 2.5	Self-healing cycles
Cost per PR	< $1.00	API + compute cost

Estimated Cost

Service	Dev/mo	Prod/mo	Enterprise/mo
Azure OpenAI	$80	$700	$2,500
GitHub Actions	$0	$80	$300
Azure Container Apps	$10	$100	$400
Azure Blob Storage	$5	$30	$100
Cosmos DB	$5	$75	$350
Key Vault	$1	$5	$15
Application Insights	$0	$30	$100
Total	$101	$1,020	$3,765

Estimates based on Azure retail pricing. Actual costs vary by region, usage, and enterprise agreements.

💰 Full cost breakdown

WAF Alignment

Pillar	Implementation
Reliability	Self-healing CI (3 attempts), existing test preservation, scope guards
Security	GitHub PAT in Key Vault, no credential in code, scope-limited PRs
Cost Optimization	gpt-4o-mini for tests+PR descriptions, cached codebase index
Operational Excellence	Webhook automation, PR labels, auto-assignment to reviewer
Performance Efficiency	One file at a time with verification, parallel test execution
Responsible AI	Max 10 files/PR scope guard, human approval before merge

FAI Manifest

Field	Value
Play	`51-autonomous-coding-agent`
Version	`1.0.0`
Knowledge	O2-Agent-Coding, F4-GitHub-Agentic-OS, T3-Production-Patterns, O3-MCP-Tools-Functions
WAF Pillars	security, reliability, operational-excellence, cost-optimization
Groundedness	≥ 85%
Safety	0 violations max