Pre-commit code review, security scanning, AI-slop scoring, change descriptions, questions, task verification, pull-request hygiene checks, docstring and unit-test generation, issue explanation, tracking-issue creation, and fix planning for Claude Code.
| Command | Purpose |
|---|---|
/cloudaeye:init |
Detect and initialize the current repository |
/cloudaeye:inspect |
Bug-focused review without security prompts |
/cloudaeye:security |
Application, LLM, agent, MCP, and secret security review |
/cloudaeye:review |
Full bug and security review |
/cloudaeye:describe |
Describe the pending change |
/cloudaeye:control-flow |
Draw what the change did to the flow of control, as a sequence diagram |
/cloudaeye:ask |
Ask a question about the pending change |
/cloudaeye:check-task |
Compare the pending change with a task or ticket |
/cloudaeye:check-pr |
Run the hygiene checklist over an open pull request |
/cloudaeye:implement |
Plan a fix for findings a review already produced |
/cloudaeye:slop-score |
Score a change for AI-slop signals, with the evidence behind every point |
/cloudaeye:add-docs |
Write docstrings for an open pull request and post them there as suggestions |
/cloudaeye:add-tests |
Write unit tests for an open pull request and post them there as suggestions |
/cloudaeye:explain |
Explain a Jira or GitHub issue against the code it touches |
/cloudaeye:create-issue |
File findings from a review as Jira tickets or GitHub issues |
The review commands report results and do not edit code. /cloudaeye:implement is the one that leads to edits, and even there the server only returns a plan — Claude applies it, and you re-run the review to confirm the fix landed.
Three commands write where other people can see it, and each confirms before it does: /cloudaeye:add-docs and /cloudaeye:add-tests post suggestions on a pull request, and /cloudaeye:create-issue opens real tickets.
/plugin marketplace add CloudAEye/claude-plugin
/plugin install cloudaeye
Restart Claude Code so the MCP server connects. If you previously installed the skills manually, remove those copies to avoid loading stale commands.
- Run
/mcpin Claude Code. - Select CloudAEye and choose Authenticate.
- Complete sign-in, organisation selection, and consent in the browser.
- Run any
/cloudaeye:*command in a Git repository with pending changes.
Claude Code stores and refreshes the OAuth credentials.
The plugin uses https://api.cloudaeye.com/mcp by default. Set CLOUDAEYE_URL before starting Claude Code to use a self-hosted OAuth-enabled MCP endpoint.
Each operational skill:
- Detects the provider, repository URL, current branch, and base branch.
- Calls the session-free
initialize_repositorytool; it opens setup only when the provider is not connected. - Calls the OAuth-authenticated
start_sessiontool after initialization. - Builds and uploads the diff with
git diffaftergit add --intent-to-add .so untracked files are included. - Calls the requested MCP review tool with the returned session ID.
The upload token is placed in a private temporary curl config, is not printed, and is deleted when the upload command exits. The .cloudaeye/ scratch directory contains only gitignored session diff data. Previously stored credential files are neither read nor changed.
OAuth requires an interactive MCP client, so unattended CI and service-account support is outside this release.
| Symptom | Fix |
|---|---|
| CloudAEye shows Needs authentication | Open /mcp and complete Authenticate |
A skill says start_session is unavailable |
Restart Claude Code so the updated MCP tools load |
upload_http=401 |
The upload grant is missing, invalid, or belongs to another session |
upload_http=000 |
Check network access and the configured review server URL |
cloudaeye_error=insecure_url |
Use HTTPS, except for localhost development |
base_source=head |
Connect the repository integration to enable the configured target branch |
.claude-plugin/plugin.json plugin manifest
.mcp.json OAuth MCP server registration
skills/<verb>/SKILL.md operational commands