developerz-ai

End-to-end PR loop. Opens PR, waits CI, waits CodeRabbit, triages, fixes, replies, resolves, re-runs CI, merges. The mastermind invokes this and walks away. Triggers: babysit pr, drive pr, ship pr end to end.

Branch naming, base branch, draft policy, auto-merge policy, force-push policy. Short, table-driven. Triggers: branch naming, branch policy, where to branch from.

When and how to evolve .coderabbit.yaml — enable new tools, switch profile, add path-instructions for new doc subtrees, tune tone. Triggers: update coderabbit config, tune cr, edit coderabbit yaml.

Post threaded replies to CodeRabbit review threads via GraphQL addPullRequestReviewThreadReply (or REST fallback). For wrong/opinion threads include principle citation. Triggers: reply coderabbit, post cr reply.

Mark review threads resolved via GraphQL resolveReviewThread mutation. Paginated, dry-run, batch. REST cannot do this. Triggers: resolve cr threads, mark resolved, close review threads.

Fetch all coderabbitai[bot] review threads via GraphQL, classify each (bug/nit/suggestion/opinion/wrong), emit accept/reject/discuss plan. Triggers: triage coderabbit, classify cr comments, cr plan.

For each accepted CR thread, locate file, route to right subagent, apply fix, build, run scoped tests, push, then resolve threads. Triggers: fix cr, apply coderabbit, address review comments.

Reusable gh api graphql snippets — list PR threads, list bot comments, resolve thread, add reply, enable auto-merge, request reviewer, dismiss review. Single source of truth for other skills. Triggers: graphql gh, gh api recipes, github graphql.

Implement or modify the in-browser live-edit engine — DOM mutation primitives, the element picker, stable-selector heuristics, and the changeset recorder. Use when touching src/dom, src/entrypoints/content.ts, or src/shared/changeset.ts, or when the task mentions live editing, DOM tools, selectors, or recording changes.

Manifest V3 extension gotchas — the three execution worlds, key custody, CSP / no-remote-code, the ephemeral service worker, and the typed message bus. Use when touching the manifest, wxt.config.ts, entrypoints, cross-world messaging, or when something behaves differently in the extension than in tests.

Assemble an accepted changeset and dispatch the implementation handoff over MCP to ai-dev / developerz.ai. Use when touching src/mcp, the ship flow in src/entrypoints/background.ts, or when the task mentions handoff, shipping a design, MCP task creation, or turning edits into a PR.

SolidJS + single-responsibility conventions for the side-panel UI — one component per file with co-located SCSS, thin stores, no logic in components. Use when creating or editing anything under src/entrypoints/sidepanel, Solid components, signals/stores, or SCSS.

Testing patterns for the extension — Vitest unit + integration projects and Playwright E2E on a loaded extension. Use when writing or fixing tests, setting up mocks for the agent loop / DOM tools / MCP, or when CI test jobs fail.

Review code for quality and security

Deploy applications to cloud environments

Review code for quality and security

Deploy applications to cloud environments

Dispatch and track goals across the fleet — use when asked to send work to a team, check goal progress, or verify a deploy.

Add one or more VPSes to the fleet and stand up their dev environment — scaffold the team yml, bootstrap tools, optimize the box, wire per-org secrets, hand off login. Use when asked to add a server/box/team, grow the fleet, or set up a dev environment on a VPS.

Operate the VPS fleet — status, accounts, usage, ssh, exec. Use when asked about servers, teams, Claude logins, or subscription usage.

Merge a GitHub PR by looping until it's green and mergeable — waits for CI, fixes failures with the Claude SDK, addresses CodeRabbit review comments, resolves conflicts, then merges. Use when the user asks to merge a PR, land a branch, get a PR green, or clear CI/review comments. Wraps `claudetm merge-pr`.

How to drive `claude` (Claude Code) as a non-interactive CLI tool — headless runs, MCP servers, permissions, output formats, and piping. Use when scripting Claude, spawning a sub-agent from another process, or wiring Claude into CI/automation.

Use whenever the user wants to run work on a remote host through an mcp-ssh server — shell commands, reading/searching/editing files, launching long-running tasks and polling them, or kicking off autonomous coding tools (if installed) such as a multi-session orchestrator or a one-shot `claude -p`. Triggers on "run on the server/host", "do it on the VPS", "kick off the agent", "start the orchestrator", "use claude -p", "check progress", "is it still running", or any goal where work should happen on the remote host rather than inline.