← All guides

Image enhancement and restoration skills for your assistant

How to install and use the CraterView skills in Claude Code and similar agents.

Last updated 17 September 2026

A skill is a folder with a SKILL.md in it: instructions an agent reads when a task matches its description, and puts down again when it does not. CraterView publishes two, as Claude Code plugins and as plain skills that work anywhere the format is read. One is for an assistant that should run the work on your account; the other is for an assistant that should write code against the API.

Install

Both come from the same marketplace, which rides on the client repositories:

/plugin marketplace add craterviewai/craterview-python
/plugin install craterview@craterviewai          # the hosted tools, and a skill for using them
/plugin install craterview-api@craterviewai      # the API, from code

(craterviewai/craterview-ts carries the same marketplace; add whichever you already have.) Once installed, the skills are /craterview:enhance-image and /craterview-api:enhance-image, and an agent also picks them up on its own when a request matches.

Without Claude Code: each plugin’s skills/enhance-image/ folder is a skill in the open format. Copy it into ~/.claude/skills/ for a personal install, or zip the folder and upload it as a custom skill to Claude.

craterview — the assistant runs the work

Installing this plugin connects CraterView’s MCP server, so the assistant can enhance, enlarge and restore photographs in the conversation. The first use opens an authorization page that asks for a code from your dashboard — Connect an assistant → Get a code — and no key is ever on screen. Connect an assistant walks through that, and through what the two tools do.

The skill beside the server is what makes the assistant good at using it. It says which model suits a soft scan and which a torn print, that a result is an expiring link rather than the bytes, that a job still running is collected rather than re-run, and what a batch will cost before it starts.

craterview-api — the assistant writes the code

No hosted tools. This skill is for an agent writing or running code against the API: a script over a directory, a service that enhances uploads, a batch with callbacks. It covers, in order:

1. The key, first. Every call needs one, so the skill checks for it before writing a line of code, and walks the user through the dashboard — Developer API → Show my API key → Copy — rather than guessing or stubbing. It asks for the key in the shell’s environment, never in the conversation:

bash export CV_API_KEY=cv_...

A key pasted into a chat is in the transcript; the skill says so and suggests rotating it afterwards.

1. The two clients, and the raw calls. pip install craterview, npm install craterview, and the three requests underneath them — ask for somewhere to put the file, upload the bytes, submit the job — for anything that is neither Python nor TypeScript.

2. Which model, and what each takes. A table of the models in service when the skill was written, with the catalogue named as the authority over it — an agent reads the live list, parameters and prices from the API when the answer matters.

3. Retries that do not double-bill. An idempotency key on the submit, so a request repeated after a dropped connection returns the original job rather than a second charge.

4. Callbacks. A webhook_url on the job, and the signature to verify on every delivery against the raw bytes.

5. Cost. Prices are per model and published by the catalogue; the skill reads them from there rather than restating them, and says what a batch will cost before running one.

It also ships a script, scripts/enhance.py, for running one image from the command line — useful as a smoke test before writing anything. Point it at echo first: that model costs nothing, needs no GPU and returns a real result of the right shape, so the plumbing is proved for free and switched to a paid model by changing one string, exactly as your first API call does:

pip install craterview
export CV_API_KEY=cv_...
python scripts/enhance.py photo.jpg --model echo --scale 2 -o photo-2x.jpg
python scripts/enhance.py scan.jpg --model cv-restore-v1 --param mode=full

The script reads the key from the environment and nowhere else — a key on a command line ends up in shell history — exits non-zero on anything the API refused, and prints a sentence saying why.

Which one

The tools route runs on the account you connected; the code route runs on whichever key the code holds. Several keys on one account is the ordinary arrangement — one per project or environment — and they share the account’s credit and history.

Two things to know

The catalogue is authoritative. A skill was written against the models in service on a given day; the API publishes the current list, each model’s parameters and its price, and an agent should read them from there. If a skill and the catalogue disagree, the catalogue is right.

Skills are versioned with the clients. They are generated into the client repositories with each release, so the skill you install matches the client it sits beside. Anything a skill gets wrong is an issue on that repository.

Next

Connect an assistant for the hosted route end to end, or your first API call to make the three requests yourself before handing them to an agent.