Context Continuity Kit

A portable memory layer for AI agents.

Context Continuity Kit is an open, filesystem-first memory system for AI work. It packages durable context into readable files instead of leaving it trapped inside a single platform's hidden memory layer.

The project ships as a Codex skill named portable-ai-memory, with a small standard-library Python CLI for creating, validating, indexing, checkpointing, and snapshotting portable AI memory packs.

GitHub Metadata

Use these values when publishing the repository:

Field	Value
Repository name	portable-ai-memory
Display name	Context Continuity Kit
Short tagline	A portable memory layer for AI agents.
GitHub About	Codex skill and CLI for portable AI memory: platform bridges, personal context, workflow skills, checkpoints, and snapshots.
Primary language	Python
License	MIT
Skill ID	portable-ai-memory
Default memory pack	ai-memory-pack

Suggested GitHub topics:

ai-memory
ai-agents
codex
claude
chatgpt
llm
context-engineering
agent-memory
workflow-automation
personal-knowledge-management

Why It Exists

AI memory is becoming infrastructure. When memory is incomplete, wrong, or locked inside one application, every new tool migration becomes a reconstruction project: refeeding context, rebuilding preferences, recovering project state, and retraining the assistant's working style.

Context Continuity Kit treats memory as a portable operating surface:

• Platform memory remains mirrored, not blindly trusted.
• Personal guidance stays user-owned.
• Project and resource maps become explicit.
• Repeated workflows become atomic skills.
• Important states can be checkpointed and snapshotted.

The result is a plug-in memory layer that can travel across Codex, Claude, ChatGPT, Gemini, and other AI tools.

What It Provides

• Platform bridges for Codex, Claude, ChatGPT, Gemini, and generic AI tools.
• Personal global guidance through GLOBAL_GUIDE.md, STATE.md, PROJECTS.md, and RESOURCES_AND_APIS.md.
• Workflow skill management for turning repeated tasks into atomic, updatable procedures.
• Deterministic maintenance through memory_pack.py.
• Checkpointing and snapshots for freezing stable state at important moments.
• Privacy boundaries that keep real memory packs, exports, logs, snapshots, and secrets out of version control.

Repository Layout

portable-ai-memory/
|-- SKILL.md
|-- agents/openai.yaml
|-- scripts/memory_pack.py
|-- references/
`-- assets/memory-pack-template/

Your real memory pack should remain local and private:

ai-memory-pack/

Do not publish real memory packs, conversation exports, snapshots, logs, credentials, or platform-specific private memory.

Install

Use the skill directly from this repository, or copy portable-ai-memory/ into your Codex skills directory:

cp -R portable-ai-memory "${CODEX_HOME:-$HOME/.codex}/skills/"

Then invoke it as $portable-ai-memory in Codex.

Quick Start

Create a local memory pack:

python3 portable-ai-memory/scripts/memory_pack.py init ./ai-memory-pack --owner "your-name"

Validate it:

python3 portable-ai-memory/scripts/memory_pack.py check ./ai-memory-pack

Refresh its index:

python3 portable-ai-memory/scripts/memory_pack.py index ./ai-memory-pack

Append a platform checkpoint:

python3 portable-ai-memory/scripts/memory_pack.py checkpoint ./ai-memory-pack \
  --platform codex \
  --title "Task complete" \
  --summary "Stable context summary for the next AI session."

Create a snapshot:

python3 portable-ai-memory/scripts/memory_pack.py snapshot ./ai-memory-pack --reason "task-complete"

Create an atomic workflow:

python3 portable-ai-memory/scripts/memory_pack.py new-workflow ./ai-memory-pack "weekly review" \
  --trigger "weekly review" \
  --goal "Summarize active projects and refresh next actions."

Memory Pack Load Order

For a new AI session, read files in this order:

1. PACK_INDEX.md
2. personal/GLOBAL_GUIDE.md
3. personal/STATE.md
4. personal/PROJECTS.md when project context is needed
5. personal/RESOURCES_AND_APIS.md when resources or APIs are needed
6. official-memory/<platform>/memory.md for platform-specific context
7. workflow-skills/<name>/SKILL.md when a workflow trigger matches

Safety Model

• Keep platform memory mirrors separate from user-owned guidance.
• Store raw conversation exports only when the user intentionally provides them.
• Keep secrets out of memory files.
• Store credential references, not credential values.
• Prefer stable, actionable summaries over transcripts.
• Snapshot before risky reorganizations and after important checkpoints.

Development Checks

Run these before publishing changes:

PYTHONPYCACHEPREFIX=/tmp/portable-ai-memory-pycache python3 -m py_compile portable-ai-memory/scripts/memory_pack.py
python3 portable-ai-memory/scripts/memory_pack.py init /tmp/portable-ai-memory-test --owner tester
python3 portable-ai-memory/scripts/memory_pack.py check /tmp/portable-ai-memory-test

If your environment has PyYAML installed, you can also run Codex's skill validator:

python3 /path/to/skill-creator/scripts/quick_validate.py portable-ai-memory

Naming System

Name	Purpose
Context Continuity Kit	Public project name
portable-ai-memory	Repository name and Codex skill ID
memory_pack.py	Local maintenance CLI
ai-memory-pack	Default private memory pack folder
official-memory/	Platform-owned memory mirrors
personal/	User-owned durable guidance
workflow-skills/	Atomic repeatable workflows

License

MIT. See LICENSE.