custom-engine-implementation

A guide to building custom agent engines for GitHub Agentic Workflows, a system for running software agents from GitHub workflows. It explains the engine interfaces, optional capabilities, implementation patterns, tests, and integration checks.

In plain words
What is it for?
Designing engine interfaces, implementing custom engines, checking support for tools or web features, refactoring engine code, and planning tests and integration work.
Why use it?
It shows how to add an engine that fits the existing architecture while preserving compatibility and declaring which features it supports.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/github/gh-aw/custom-engine-implementation
Clone the repo
git clone --depth 1 https://github.com/github/gh-aw
Per session 12 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 6,106 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00012 $0.06106
Opus 5 $0.00006 $0.03053
Sonnet 5 $0.00002 $0.01221
Haiku 4.5 $0.00001 $0.00611

Measured 2d ago against content hash 635973fb1c27, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

custom-engine-implementation scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.github/agents/custom-engine-implementation.agent.md · 826 lines

How it starts

The opening of the file, as written. The whole thing — 826 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Custom Agentic Engine Implementation Guide

This document provides a comprehensive guide for implementing custom agentic engines in GitHub Agentic Workflows (gh-aw). It covers architecture patterns, common refactoring opportunities, and step-by-step implementation instructions.

Table of Contents

  1. Architecture Overview
  2. Engine Interface Design
  3. Common Code Analysis & Refactoring Opportunities
  4. Implementation Guide
  5. Testing Strategy
  6. Integration Checklist

Architecture Overview

Interface Segregation Principle

The agentic engine architecture follows the Interface Segregation Principle (ISP) to avoid forcing implementations to depend on methods they don't use. The system uses interface composition to provide flexibility while maintaining backward compatibility.

Interface Hierarchy

Engine (core identity - required by all)
├── GetID()
├── GetDisplayName()
├── GetDescription()
└── IsExperimental()

CapabilityProvider (feature detection - optional)
├── SupportsToolsAllowlist()
├── SupportsHTTPTransport()
├── SupportsMaxTurns()
├── SupportsWebFetch()
├── SupportsWebSearch()
├── SupportsFirewall()
├── SupportsPlugins()
└── SupportsLLMGateway()

WorkflowExecutor (compilation - required)
├── GetDeclaredOutputFiles()
├── GetInstallationSteps()
└── GetExecutionSteps()

MCPConfigProvider (MCP servers - optional)
└── RenderMCPConfig()

LogParser (log analysis - optional)
├── ParseLogMetrics()
├── GetLogParserScriptId()
└── GetLogFileForParsing()

SecurityProvider (security features - optional)
├── GetDefaultDetectionModel()
└── GetRequiredSecretNames()

CodingAgentEngine (composite - backward compatibility)
└── Composes all above interfaces

Key Architectural Patterns

  1. BaseEngine Embedding: All engines embed BaseEngine which provides default implementations
  2. Focused Interfaces: Each interface has a single responsibility
  3. Optional Capabilities: Engines override only the methods they need
  4. Backward Compatibility: CodingAgentEngine composite interface maintains compatibility

Read the full file on GitHub · 826 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 826 lines · 12 tokens per session scan A 635973fb1c27

Subscribe to this mod's changes

custom-engine-implementation is an agent published in the GitHub repository github/gh-aw (5,050 stars, last pushed 2d ago), licensed MIT. It adds 12 tokens to every session and 6,106 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other agents, from other repositories

devops-engineer

CI/CD, deployment, and infrastructure automation specialist.

ChanningLua/prax-agent · 14 tokens

checker-engineer

Use when a diff, planned change, or OpenSpec proposal touches the checker kernel — packages/claims/src/checkClaims.ts, witness.ts, wiring.ts, rules.ts, or config.ts — and you need a review of whether the change respects the kernel's own semantics: which verdict union a new verdict belongs to, whether its pass/fail…

armanfatemi/nullius · 275 tokens

retro-writer

Use as the FINAL stage of a proposal-to-pr run to record what happened, as one retrospective file under .claude/retrospectives/. Dispatched fresh, having NOT done the work, so it reads artefacts — the pipeline state file, review-evidence.md and its ## Probe — stage 2 score, progress.md, git history — rather than the…

armanfatemi/nullius · 252 tokens

github-actions-expert

Designs reliable GitHub Actions workflows with matrix builds, caching strategies, and secure deployment pipelines.

NickCrew/Claude-Cortex · 24 tokens

present-agent

An agent that exists, so the skill dispatching to it resolves.

armanfatemi/nullius · 2 tokens

bolt

Learning: Checking a file path against multiple GlobSets sequentially is less efficient than combining them into a single GlobSet and checking match indices. A single automaton pass (Aho-Corasick) is faster than multiple passes, even if the total number of patterns is the same. Action: When classifying strings against…

EffortlessMetrics/xchecker · 0 tokens