Agent Foundry

agent-foundry is a shared repository for future agents, skills, and MCP integrations designed to support workflows across tools such as Codex and Claude Code.

This repository is intended to become a central, curated source of reusable AI building blocks, including:

• agent definitions
• skills and workflows
• MCP artifacts, launchers, and templates
• governance and provenance metadata
• licensing and attribution records

Status

This repository is currently in its initial setup phase.

At this stage, it contains:

• repository governance files
• licensing and notice files
• initial catalog metadata
• supporting documentation
• initial native skill definitions
• Codex subagents under agents/
• initial MCP artifact definitions under mcps/

Goals

The main goals of this repository are:

• provide a structured home for reusable AI development assets
• support both native and adapted artifacts
• keep provenance and licensing clear
• make future imports and adaptations easier to track
• establish a clean foundation before adding operational content

Repository Structure

Current structure:

agent-foundry/
├─ LICENSE
├─ NOTICE
├─ README.md
├─ registry.yaml
├─ agents/
├─ docs/
├─ mcps/
├─ scripts/
├─ skills/
└─ third_party/
   ├─ THIRD_PARTY_NOTICES.md
   └─ licenses/

Planned future areas may include:

catalog/
templates/
hooks/

Licensing

This repository is licensed under Apache License 2.0 for original content unless stated otherwise.

Third-party materials, when added, will retain their original license terms and required notices.
Attribution and provenance will be documented in the appropriate notice and registry files.

Provenance and Attribution

This repository is being designed with provenance tracking in mind.

When third-party content is imported or adapted, the project aims to document:

• original source repository
• original license
• modification status
• ownership and maintenance metadata

Registry

The registry.yaml file is intended to act as the base catalog for repository contents.

Over time, it may track:

• agents
• skills
• MCP integrations
• ownership
• status
• licensing metadata
• provenance details

Installing Codex Assets

Use the installer to synchronize this repository's Codex agents and skills into your Codex home directory:

python3 scripts/install_foundry.py

By default, the script installs agents into ~/.codex/agents and skills into ~/.codex/skills, using CODEX_HOME when that environment variable is set. It is idempotent: existing files are skipped when identical and updated only when the repository copy differs.

Useful maintenance commands:

# Preview install or update changes without writing files.
python3 scripts/install_foundry.py --dry-run

# CI/check mode: exit 1 when installed assets are stale or missing.
python3 scripts/install_foundry.py --check

# Install only one component.
python3 scripts/install_foundry.py --component agents
python3 scripts/install_foundry.py --component skills

# Remove files previously installed by this script when their source is gone.
python3 scripts/install_foundry.py --prune

The installer writes a manifest at $CODEX_HOME/.agent-foundry-install.json. Pruning uses that manifest and leaves locally modified stale files in place.

Documentation

The docs/ directory contains supporting research and reference material relevant to the repository’s long-term direction.

These documents help define the architectural and operational basis for future additions.

Third-Party Notices

The third_party/ directory is reserved for third-party attribution and notice records.

As external materials are added in the future, this area will be used to keep licensing and attribution clear and organized. For shared imported or adapted agent collections, upstream license texts belong under third_party/licenses/; do not duplicate folder-level license files under agents/ unless a future source specifically requires local packaging.

Roadmap

Planned next steps include:

1. define the initial catalog model
2. establish import and adaptation rules
3. curate imported agents and add the first native or adapted agents
4. expand the initial skill definitions
5. expand MCP integration templates and installation helpers
6. introduce validation and maintenance workflows

Contribution Philosophy

This repository is intended to prioritize:

• clarity
• traceability
• maintainability
• compatibility across tools
• responsible reuse of external artifacts