Copilot Agent Settings

Shared config and templates for AI Software House Copilot CLI sessions.

This repository is the single source of truth for agent role prompts, memory bank templates, and helper scripts used by the copilot-software-house and ai-software-house orchestration pipelines.

---

What's in here

Path	Purpose
.github/copilot-instructions.md	Auto-loaded by Copilot CLI; tells it to read the Memory Bank at session start
memory-bank/	6 Markdown template files that give Copilot persistent memory across sessions
agents/	13 system-prompt files — one per pipeline agent role
skills/	Python tool files (builtin_tools.py, tool_registry.py) for agent capabilities
deploy-memory-bank.sh	Copies memory-bank/ and copilot-instructions.md into any target project
update-memory-bank.sh	Manual memory bank update triggered via Copilot CLI
install-memory-bank-hook.sh	Installs a git post-commit hook that semi-automatically updates the memory bank
docs/superpowers/	Design specs and plans for the superpowers skill system

---

Memory Bank

The Memory Bank is the most important concept in this repo. It gives Copilot persistent, structured context across sessions — so it always knows what the project is, how it is architected, and where work currently stands.

It is inspired by Cline's Memory Bank pattern, adapted for Copilot CLI.

The 6 files and their hierarchy

The files form a deliberate reading order from stable context to current state:

projectbrief.md          ← why the project exists; goals and scope
  └── productContext.md  ← user problems, UX goals, success criteria
        ├── systemPatterns.md  ← architecture, patterns, conventions
        ├── techContext.md     ← tech stack, dependencies, environment
        └── activeContext.md   ← current focus, recent changes, next steps
              └── progress.md  ← what's done, in-progress, blocked

projectbrief.md and productContext.md are filled in once and rarely change. activeContext.md and progress.md are updated after every significant session.

Setup — deploy to a project

Run this once per project:

cd /path/to/copilot-agent-setting
./deploy-memory-bank.sh /path/to/your-project

This copies the 6 template files into your-project/memory-bank/ and writes .github/copilot-instructions.md so Copilot reads them automatically.

Then edit the two foundation files for your project:

# Fill these in before starting any work
your-project/memory-bank/projectbrief.md
your-project/memory-bank/productContext.md

Three update modes

Fully automatic

Both copilot-software-house and ai-software-house orchestrators automatically call the memory_bank_updater agent after every pipeline run, keeping all 6 files up to date without any manual steps.

Semi-automatic (git hook)

Install once per project. After that, every git commit automatically updates activeContext.md and progress.md:

cd /path/to/copilot-agent-setting
./install-memory-bank-hook.sh /path/to/your-project

Manual

Run from inside your project directory, passing a short description of what changed:

./update-memory-bank.sh "Added JWT authentication to /api/auth"

How Copilot reads it

deploy-memory-bank.sh writes .github/copilot-instructions.md into your project. Copilot CLI auto-loads that file at the start of every session. It instructs Copilot to read all 6 memory bank files in order before doing anything else, so full project context is always available from the first message.

---

Agent Roles

The agents/ directory contains 13 system-prompt files. Each file defines the persona, responsibilities, and output format for one stage of the AI Software House pipeline.

File	Role
product_manager.md	Turns raw requirements into a structured PRD with user stories and acceptance criteria
pm_reviewer.md	Reviews PRDs for completeness, clarity, and testability before design begins
architect.md	Produces system design (data models, API contracts, module breakdown) from a PRD
architect_reviewer.md	Reviews designs for completeness, correctness, and feasibility
engineer.md	Implements modules from a system design, writing clean production code
code_reviewer.md	Reviews code for bugs, security vulnerabilities, and quality
qa_planner.md	Writes a comprehensive test plan from acceptance criteria
qa_engineer.md	Writes immediately runnable pytest test suites covering all layers
deployment_tester.md	Writes deployment smoke tests and docker-compose test configs
summariser.md	Writes a compact, factual memory entry after each pipeline run
memory_bank_updater.md	Updates the 6 memory bank files after each run
memory_consolidator.md	Compresses multiple run summaries into a single snapshot
refactor_agent.md	Reviews and rewrites code for readability and maintainability

---

Using with copilot-software-house / ai-software-house

Both orchestrator systems read agent role files directly from this repository. When a pipeline run kicks off, each stage loads its corresponding file from agents/ as its system prompt, so every agent operates with a consistent and well-defined persona. At the end of each run the orchestrator calls memory_bank_updater to keep the Memory Bank in the target project current — meaning the next session (human or automated) always starts with accurate context.

---

License

MIT