Project Memory

Project Memory is a reusable coding-agent skill for durable, token-efficient project continuity.

Instead of storing every implementation update in one ever-growing Markdown file, it uses a small summary plus topic-specific records:

.agents/MEMORY/
├── SUMMARY.md
└── records/
    ├── architecture/
    ├── decisions/
    ├── features/
    ├── incidents/
    └── migrations/

Agents always read SUMMARY.md and load only the records relevant to their current task.

Why

A single implementation-status document eventually becomes expensive to read, difficult to maintain, and full of stale chronology. Project Memory keeps current state concise while preserving durable detail in targeted records.

Install

Clone this repository into a project's skill directory:

git clone https://github.com/pedroantonnio/project-memory.git .agents/SKILLS/project-memory

Then require agents to read:

.agents/SKILLS/project-memory/SKILL.md

before implementation.

Quick start

python .agents/SKILLS/project-memory/scripts/memory.py --repo . init
python .agents/SKILLS/project-memory/scripts/memory.py --repo . status
python .agents/SKILLS/project-memory/scripts/memory.py --repo . create --type feature --slug finances --title "Finances"
python .agents/SKILLS/project-memory/scripts/memory.py --repo . validate

Principles

• Summary is small and current.
• Records hold durable detail.
• Only relevant records are loaded.
• Significant project knowledge is versioned with the code.
• Temporary task coordination stays in the coordination system, not Project Memory.
• No secrets or private chain-of-thought are stored.

See SKILL.md for the full protocol.