A Particular Set of Skills

Models and agents can be replaced. What they learn about their user should survive the replacement.

This is a collection of independently installable Agent Skills built on Fulcra. Each skill gives an agent a focused workflow over the owner's context lake. Everything required at runtime lives inside the skill directory.

Each skill declares whether it accesses Fulcra through the CLI or an authenticated MCP connection. Users need only the interface declared by that skill.

CLI-backed skills use the current Fulcra CLI:

uvx --from fulcra-api@latest fulcra <command>

Available skills

Skill	What it does	Compatibility
Request Image Generation	Lets an MCP-only AI product request images and retrieve verified candidates contributed by an image-capable agent.	Authenticated Fulcra MCP with read and write access
Image Upgrader	Finds Image Upgrade Requests, generates candidates, publishes verified image bytes, and records linked Contributions.	Image generation, uv, network access, authenticated Fulcra CLI, and an HTTPS publishing route
Computer History Collector	Copies completed Codex Computer History summaries into Fulcra so agents the owner authorizes can read the same work history.	macOS, Codex Computer History, uv, and network access

Request Image Generation and Image Upgrader are two sides of one Typed Blackboard workflow. A product such as Claude can write a text-only Request through Fulcra MCP. A separate agent such as Codex can generate and publish the image, then leave a digest-addressed Contribution for Claude to retrieve and render. The first release supports agents connected to the same Fulcra owner.

Repository layout

.
├── skills/
│   └── <skill-name>/
│       ├── README.md      # human-facing purpose and installation
│       ├── SKILL.md
│       ├── scripts/       # optional
│       ├── references/    # optional
│       └── assets/        # optional
├── tests/
│   └── acceptance/
│       └── <skill-name>/  # repository-only live-account tests
├── docs/                  # authoring conventions and ADRs
├── AGENTS.md
└── README.md

Using the skills

Each skill follows the Agent Skills specification and can be installed independently by a compatible agent client. Each skill directory documents its own purpose and installation.

Developing a skill

1. Clone this repository to adopt its shared authoring conventions.
2. Create the skill under skills/<skill-name>/.
3. Keep everything required at runtime inside that skill directory.
4. Validate and test the skill according to docs/agents/skill-authoring.md, including acceptance against a live Fulcra account.
5. Keep private acceptance artifacts under the ignored .acceptance-runs/ directory.
6. Distribute the individual skill directory without the repository-level development infrastructure.

Repository-wide references:

• AGENTS.md contains the repository-wide agent instructions.
• docs/agents/skill-authoring.md defines the portable skill boundary and completion criteria.
• tests/acceptance/ contains repository-only live-account acceptance scenarios.
• docs/adr/ records consequential authoring and release decisions.

This is a public repository. Raw Fulcra account data, acceptance output, cleanup manifests, credentials, and local environment files remain local and are never committed.

License

Licensed under the Apache License 2.0.