Enterprise Architect Personal Assistant — Copilot Customization

This repository (.github-private) configures GitHub Copilot as a personal enterprise architecture assistant for the organization. It provides custom agents, reusable prompts, domain skills, coding instructions, and grounding hooks — all designed to augment architectural decision-making with rigorous source citation and anti-hallucination safeguards.

Repository Structure

.github-private/
├── copilot-instructions.md          # Global Copilot behavior and rules
├── agents/                          # Custom agent definitions
│   ├── proposal-writer.agent.md     # RFP and client document drafting
│   ├── researcher.agent.md          # Deep research with citations
│   ├── reviewer.agent.md            # Code/PR review specialist
│   └── solution-architect.agent.md  # Architecture design and ADRs
├── prompts/                         # Reusable prompt templates (invoked via /)
│   ├── code-review.prompt.md        # Structured code review
│   ├── compare-solutions.prompt.md  # Technology/architecture comparison
│   ├── draft-email-response.prompt.md # Professional email drafting
│   ├── estimate-effort.prompt.md    # Effort estimation with ranges
│   └── summarize-document.prompt.md # Document parsing and summarization
├── skills/                          # Domain skill definitions
│   ├── code-review/                 # Review checklists (security, Python)
│   ├── deep-research/               # Multi-source research procedure
│   ├── presentation-builder/        # PowerPoint generation
│   ├── requirements-analysis/       # Structured requirements extraction
│   ├── rfp-proposal/                # Proposal writing with templates
│   └── solution-design/             # Architecture design with ADR templates
├── instructions/                    # Contextual coding instructions
│   ├── architecture-patterns.instructions.md
│   └── python-standards.instructions.md
├── hooks/                           # Copilot hooks for runtime enforcement
│   ├── grounding_enforcer.py        # Validates grounding before submission
│   └── grounding-enforcer.json      # Hook configuration
└── ISSUE_TEMPLATE/                  # GitHub issue templates
    ├── document-analysis.md
    ├── research-task.md
    └── solution-comparison.md

Agents

Agent	Purpose	Key Skill
Proposal Writer	Drafts RFP responses and client documents grounded in parsed requirements	rfp-proposal
Researcher	Gathers, verifies, and cites information from multiple sources	deep-research
Reviewer	Reviews Python code and PRs with security/performance checklists	code-review
Solution Architect	Designs systems, evaluates technologies, and documents ADRs	solution-design

Prompts

Invoke prompts by typing / in the Copilot chat panel:

Prompt	Routed To	Description
/code-review	Reviewer	Security, performance, and maintainability review
/compare-solutions	Solution Architect	Trade-off matrix for technology/architecture options
/draft-email-response	Proposal Writer	Professional email drafting with M365 context
/estimate-effort	Solution Architect	T-shirt sizing and effort ranges
/summarize-document	Researcher	Parse and summarize client documents

Skills

Skills provide detailed procedures and reference materials for complex tasks:

• code-review — Checklists for security (OWASP-mapped) and Python best practices
• deep-research — Multi-source research with citation tracking
• presentation-builder — Generate PowerPoint decks from content
• requirements-analysis — Extract structured requirements from documents
• rfp-proposal — Proposal generation using organizational templates
• solution-design — Architecture design with ADR documentation

Instructions

Contextual instructions are applied automatically based on file patterns or topic:

• python-standards — Applied to all **/*.py files (Python 3.11+, type hints, pathlib, logging)
• architecture-patterns — Applied when discussing system architecture or design decisions

Hooks

The grounding enforcer hook runs on every prompt submission to validate that responses follow the anti-hallucination rules defined in copilot-instructions.md.

Core Principles

1. Never fabricate information — Every claim must cite a verifiable source
2. Tools before memory — Always use parsers, WorkIQ, or web search rather than relying on training data
3. Separate fact from opinion — Explicit markers distinguish grounded facts from professional judgment
4. Flag uncertainties — Mark assumptions and unverified claims with [NEEDS REVIEW: reason]
5. Present options — Architecture decisions always include at least 2 alternatives with trade-offs

MCP Integrations

• WorkIQ (ask_work_iq) — Query M365 for emails, meetings, Teams messages, documents, people
• Tavily Web Search (tavily-search) — Search the web for technology research and vendor comparisons
• Tavily Extract (tavily-extract) — Retrieve full page content from URLs
• GitHub Copilot MCP — GitHub-provided remote tools

Getting Started

1. This repository is automatically detected by GitHub Copilot for all organization members.
2. Use / in Copilot chat to invoke prompts and skills.
3. Tag an agent by name (e.g., @Reviewer) for specialized tasks.
4. The global rules in copilot-instructions.md apply to all interactions automatically.