Chisle
코딩 에이전트 플러그인, 20개 실시간 작업에서 기본 모델 52% 활용, 간결한 페르소나와 출력 압축 제공
저장소 정보는 아직 수집하지 않았습니다.
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
Community plugin packaging the OpenArtifact MCP server for Claude Code and Cursor (read-only)
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Use your OpenArtifact registry from a coding agent. The plugin connects the agent to OpenArtifact's MCP server and teaches it five things:
| Skill | What the agent does |
|---|---|
configure-project | Points Composer, npm, pip, Go, Cargo and other package managers at your OpenArtifact repository, from the server's own setup snippets, without writing a token into the project |
check-dependency | Before adding or upgrading a dependency, checks that the registry will serve that version (curation rules, minimum age, quarantine, license policy) and picks one it will |
check-lockfile | Checks every dependency of a composer.lock or package-lock.json against the registry in one call, also through a group, and reports what is held, missing or not yet known (needs a server with the check_dependencies tool; older servers get a check of the direct dependencies) |
find-artifact | Finds packages, files and images across the repositories you can read: by name, by project, by your role in a project ("shared with me") or by property (prop:status=approved), and gives the version list and download URL |
explain-install-failure | Turns a 403 or 404 from the registry into the reason (with the rule, or when the version unlocks) and a fix that respects it |
The plugin is read-only: it only calls OpenArtifact's read tools, and the agent is told never to work around the registry (no public-registry fallback, no disabled config).
MCP_ENABLED=true, and BASE_URL set to the address clients reach). See MCP server.claude plugin marketplace add openartifact/openartifact-agent-plugin
claude plugin install openartifact@openartifact
Claude Code asks for the server URL and the token when the plugin is enabled; the token is kept in secure storage. To change them later, run /plugin configure openartifact@openartifact inside Claude Code.
The plugin also carries a Cursor manifest (.cursor-plugin/plugin.json, mcp.json). Cursor takes the server URL (OA_URL) and the token (API_TOKEN) as plugin variables, which a team administrator sets in the Cursor dashboard under Plugins → Configure.
Install from a local checkout and set the options (claude plugin configure only works on an installed plugin, so --plugin-dir alone cannot supply them):
claude plugin marketplace add .
claude plugin install openartifact@openartifact
claude mcp list # after the options are set: plugin:openartifact:openartifact ... ✔ Connected
Set the options with /plugin configure openartifact@openartifact inside Claude Code: it prompts for them and keeps the token in secure storage. claude plugin configure openartifact@openartifact on its own only lists them.
For automation, pipe the options in from variables your secret store fills, so the token never appears on a command line or in shell history:
jq -n --arg url "$OA_URL" --arg token "$OA_TOKEN" '{server_url: $url, api_token: $token}' \
| claude plugin configure openartifact@openartifact --values-stdin
Check the manifests with python3 scripts/check-plugin.py (what CI runs, with REUSE and workflow lint) and claude plugin validate .. To keep a test away from your own Claude Code setup, run these with CLAUDE_CONFIG_DIR pointing to an empty directory.
evals/ holds test prompts that run the skills in a real Claude Code session against mocked OpenArtifact tools (evals/mocks/openartifact/, no server needed) and score the answers:
| Case | Checks |
|---|---|
blocked-version | a version a curation rule blocks is refused with the reason, and never worked around |
lockfile-unknown-is-not-blocked | a lockfile is checked in one check_dependencies call, and unknown is reported as "not cached yet", not as blocked |
blocked-repository-403 | a 403 on every package from a blocked repository is not blamed on missing access |
configure-keeps-the-token-out | project setup never writes the token into a file or repeats it in a command |
claude plugin eval . --case blocked-version --runs 1 --ablation none # one case, once
claude plugin eval . --scaffold --allow-tools Edit Write # the whole suite
The last case creates a composer.json with a scaffold script and edits it, so it needs --scaffold --allow-tools Edit Write. Runs use your own Claude credentials and count against your usage: one run of one case costs about $0.10 at list price, and the default is each case's runs with and without the plugin. CI does not run them. Run the suite before a release and after changing a skill.
The plugin sends requests only to the OpenArtifact server you configure, with the token you give it. It has no server of its own, collects nothing and sends nothing anywhere else. What your OpenArtifact server logs is governed by whoever runs it.
AGPL-3.0-or-later, the license of OpenArtifact. See LICENSES.