Multi-agent code review: full-review cross-correlates several reviewers and has the finding-verifier agent check every finding; code-review-git posts line-level findings as a GitHub PR review and answers other reviewers' comments. Bundles the code-review-agent, security-auditor, code-metrics-analyst, and finding-verifier agents; code-review-git verifies every finding before it is shown. Also security-check, a focused security review of a branch, and the dependency-auditor agent.
8 skills · 3 commands · 5 agents
Cost-aware routing for Claude Code: a routing skill and session-start summary that tell the main session which tier handles which task, worker subagents pinned to Sonnet or Haiku with explicit tool lists, hooks that warn when a generic subagent is dispatched without a model, log each dispatch and permission request, and count advisor calls, plus opt-in alerts when Claude Code waits on a permission prompt, an opt-in live HTML board of a session's agents and tool calls, and a command that runs one task as a headless claude -p billed to an Anthropic API key instead of the subscription. Core-risk work and reviews stay at Opus or above; contained risk may run on Sonnet or grok and then gets the Opus review. Warns, never blocks. Requires jq. In Codex, the routing skill and seven worker procedures install without the hooks, alerts, or board.
2 skills · 6 commands · 7 agents · hooks
Skills for ThryX project management over MCP: issues, projects, cycles, the client-facing Roadmap (releases, promises, outcomes, and its public link), and documents in the ThryX workspaces you connect, plus a product-management skill for writing tickets, PRDs, ADRs, and project docs and running follow-ups.
6 skills · 3 commands
Specialist subagents that take a noisy job and return one structured report: log-analyst (correlate logs to a root cause), codebase-archaeologist (map an unfamiliar or legacy system), performance-optimizer (measure, fix by impact, re-measure), and migration-specialist (breaking-change catalog and a phased plan with rollback points).
4 skills · 4 agents
War room: multi-persona deliberation for hard-to-reverse decisions, with game-theory lenses, a red-team pass, and a portfolio of options instead of a single winner. Also prd-create (section-by-section PRDs with pressure-testing, optionally saved to ThryX as a document plus epic and stories), task-architect (turns an idea into an agent-ready prompt), and the architecture-advisor agent. Also diagram-generator: Mermaid diagrams from a description or from the codebase.
5 skills · 1 agent
Debug a running web app by driving the browser: reproduce the symptom, read console and network errors, fix, and verify in the page. Also css-alignment-debug, which injects temporary outline overlays and reads a screenshot of the rendered page for layout bugs that won't resolve from reading source. Also browser-testing: Playwright test scripts with dev-server detection; Playwright installs into the plugin's data directory during a one-time setup.
3 skills