AI Agent Prompts & Playbooks for Context Engineering

A template library of coding standards, assessment playbooks, PR review playbooks, and agent configuration files. Import into any repository to get consistent AI-assisted development across all major coding agents.

Context-Optimised Architecture

Content is organised to minimise always-in-context footprint and load detail on demand:

Tier	Directory	When loaded	Purpose
1 — Always in context	core/	Session start	Lean project config (~60 lines), context index, conventions
2 — On demand	playbooks/	When task matches keywords	Assessment, review, planning, and refactoring procedures
3 — Reference	standards/	When domain matches keywords	Detailed per-concern standards (security, testing, etc.)

Result: Always-in-context is ~60 lines of AGENTS.md plus a routing table. All detail loads on demand via keyword matching in .context/index.md.

How Context Loading Works

The .context/index.md file is a keyword-to-file routing table. Every agent — Claude Code, Cursor, Copilot, Devin, Windsurf — is instructed to read this index before starting a task and load files matching the current domain.

This is the cross-agent mechanism: any LLM-based agent can read a markdown table and match keywords. No proprietary skill system required.

For Claude Code and GitHub Copilot specifically, deploy.sh generates thin skill wrappers (.claude/skills/ and .github/skills/) for the selected agents. The playbook is the single source of truth; the skill wrapper is a disposable adapter.

Example

User says: "refactor the authentication module"

1. Agent reads .context/index.md
2. Matches keyword "refactor" → .context/playbooks/refactor/safe-refactor.md
3. Matches keyword "auth" + "security" → .context/standards/security.md
4. Loads both files and follows the playbook

Supported Agents

Agent	File(s) read	How
Devin	AGENTS.md, .devin/devin.json	Native — reads AGENTS.md + .context/index.md via instructions
Cursor	.cursor/rules/standards.mdc → AGENTS.md	Redirect with alwaysApply: true
Windsurf	.windsurfrules → AGENTS.md	Redirect
Claude Code	CLAUDE.md → AGENTS.md + .claude/skills/	Delegation + generated skill wrappers
GitHub Copilot	.github/copilot-instructions.md → AGENTS.md + .github/skills/	Delegation + generated skill wrappers

AGENTS.md is the single source of truth for project conventions. All agent config files redirect to it and to .context/index.md.

Quick Start

./deploy.sh --agents all /path/to/target-repo

Examples:

# Claude only
./deploy.sh --agents claude /path/to/target-repo

# Multiple agents
./deploy.sh --agents claude copilot cursor /path/to/target-repo

# Deploy to current directory
./deploy.sh --agents all

# Omit --agents to use the interactive arrow-key selector
./deploy.sh /path/to/target-repo

Then fill in all [CONFIGURE] sections in AGENTS.md and any selected agent-specific files (for example CLAUDE.md when Claude is selected).

Repository Structure

core/                                   Tier 1 — always in context (→ target repo root)
  AGENTS.md                             Lean project config (~60 lines)
  CLAUDE.md                             Claude Code config (→ AGENTS.md + index)
  .context/
    index.md                            Keyword → file routing table
    conventions/
      code.md                           Naming, patterns, imports, core principles
      workflow.md                       Workflow orchestration, task management
      communication.md                  Writing standards, communication style
  .windsurfrules                        Windsurf redirect → AGENTS.md
  .cursor/rules/standards.mdc           Cursor redirect → AGENTS.md + index
  .devin/devin.json                     Devin config + index pointer
  .github/copilot-instructions.md       Copilot redirect → AGENTS.md
  .claude/settings.json                 Claude Code permissions + hooks template
                                        (skill wrappers are generated by deploy.sh
                                        for selected Claude/Copilot agents)

standards/                              Tier 3 — reference (→ target .context/standards/)
  code-quality.md                       SOLID, DRY, Clean Code, Clean Architecture
  security.md                           OWASP Top 10 security checklist
  testing.md                            Test Trophy Model, coverage, fixtures
  ci-cd.md                              CI/CD pipeline and quality gates
  observability.md                      OpenTelemetry, Golden Signals, 3 pillars
  resilience.md                         Circuit breakers, retries, bulkheads
  performance.md                        Performance and scalability patterns
  cost-optimisation.md                  FinOps and cost-aware engineering
  operational-excellence.md             Runbooks, config, change management
  api-design.md                         REST/GraphQL API design standards
  aws-well-architected.md               AWS Well-Architected Framework (6 pillars)
  azure-well-architected.md             Azure Well-Architected Framework (5 pillars)
  gdpr.md                               GDPR data protection standards
  pci-dss.md                            PCI DSS payment card data standards
  accessibility.md                      WCAG 2.2 Level AA accessibility standards
  architecture.md                       Clean Architecture, dependency direction, layers
  iac.md                                Infrastructure as Code, state, containers, policy
  tech-debt.md                          Debt taxonomy, impact scoring, paydown strategy
  dotnet.md                             C#, ASP.NET Core, EF Core, async patterns
  react.md                              Component architecture, hooks, Testing Library
  mssql.md                              Schema design, queries, migrations, Azure SQL
  powershell.md                         Verb-Noun, parameters, Pester, Az module
  terraform.md                          HCL file layout, modules, tflint, Terratest
  ado-pipelines.md                      YAML triggers, templates, environments, approvals
  docker.md                             Multi-stage builds, layer optimisation, scanning

playbooks/                              Tier 2 — on demand (→ target .context/playbooks/)
  assess/                               Structured codebase-level assessments (14)
    accessibility.md, api-design.md, architecture.md, aws-well-architected.md,
    azure-well-architected.md, code-quality.md, compliance.md, full.md,
    iac.md, observability.md, performance.md, security.md, tech-debt.md,
    test-coverage.md
  review/                               PR-level and change-level reviews (10)
    accessibility.md, api-design.md, architecture.md, code-quality.md,
    compliance.md, iac.md, observability.md, performance.md,
    security.md, test-coverage.md
  plan/                                 Design and decision documents (4)
    adr.md, design-doc.md, risk-assessment.md, spike.md
  refactor/                             Structured code change procedures (3)
    safe-refactor.md, extract-module.md, dependency-upgrade.md

Playbook Format

Playbooks use a universal markdown format with YAML frontmatter:

---
name: assess-security
description: "Run comprehensive OWASP Top 10 security assessment..."
keywords: [assess security, security audit, threat model]
---

# Security Assessment

## Phase 1: Discovery
...

The keywords field feeds the context index. The description field is used by deploy.sh to generate Claude Code and GitHub Copilot skill wrappers. The content is plain markdown that any agent can read and follow.

Conventions

• British English throughout (optimisation, behaviour, colour)
• Kebab-case file names
• In standards/, use ## N · Section Title heading style (middle dot separator)
• In playbooks/, use semantic non-numbered headings such as ## Role and ## Phase 1: Discovery
• [CONFIGURE] in headings marks project-specific sections
• <!-- PROJECT: ... --> HTML comments mark inline customisation points
• SOLID principles always use full names (never SRP, OCP, etc.)
• All instructions prescriptive: "must", "never", "always"

Updating Standards

Standards are maintained in one place only:

What	Where
Project conventions and workflow	core/AGENTS.md
Per-concern detail	standards/{concern}.md
On-demand context routing	core/.context/index.md
Playbook procedures	playbooks/{category}/{concern}.md
Skill wrappers (Claude + Copilot)	Generated by deploy.sh — do not edit directly

Editing Guidelines

• Do not modify templates unless explicitly asked.
• Keep standards prescriptive — "must", "never", "always" — not advisory.
• Every instruction file ends with ## Non-Negotiables and ## Decision Checklist.
• When adding a new standard, add it to standards/, create playbooks in playbooks/, and add entries to core/.context/index.md.