Virtual dev team for multi-repo legacy work - one interactive session owns a ticket from brainstorm through spec, plan, implement, tests, docs and review to done; 4 context-isolated agents (explorer, reviewer, security, researcher), a Haiku 5.5 session-relay transport agent, plus on-demand stack skills, 41 slash commands, 33 bundled skills, and deterministic hooks: a budgeted context hook (on by default) that carries the handoff across a clear or compaction, plan-approval and scope guards with a completion audit, an opt-in cloud/destructive guard, and a verify gate that fails unverified turns. /crew:migrate carries a 0.20 setup forward.
33 skills · 41 commands · 5 agents · hooks
Nuclei vulnerability scans on hosts and websites, or the whole scanner routine (checkov, trivy, dependency-check, semgrep, ZAP, testssl, nmap, nikto, gated sqlmap) from one manifest with a coverage table and a dated docs/security-scans layout, diffed against a baseline, rendered as triaged Markdown/HTML/PDF reports that itemise Critical/High/Medium and count the rest, and turned into ServiceDesk Plus tickets for Critical and High findings after one batch confirmation. No hooks, no agents.
1 skill · 6 commands
Local models on your own GPU via Ollama, in two halves - three MCP tools that index a repo or folder into a vector store on disk, search it, and answer from it, plus a `localgpu` command that opens a separate Claude Code session on a local chat model through a translating proxy. Commands for setup, health checks, indexing, search, ask, and pointing crew at a local model. No hooks and no agents; no prompt, file, or embedding leaves the machine.
1 skill · 6 commands
Two adversarial reviewers from different model families - a Claude agent and a Codex agent - tear apart a Claude Code agent, skill or plugin against one shared rubric and return a viability verdict with cited defects, weaknesses and the edits worth making. Use when reviewing a skill, agent, command or plugin as an artifact: 'is this skill worth shipping', 'tear apart this plugin', 'will this agent's description actually trigger', 'review my SKILL.md', 'does this duplicate something we already have'. Coverage is a reported outcome: if only one reviewer runs, or the two models cannot be resolved to different families, the report says so in its own title rather than passing a single-family opinion off as two. Do NOT use for architecture, implementation plans, product ideas or code diffs - use crew's architect-reviewer, planner or code-reviewer. Standalone - its own two-model config, never crew's. No hooks.
2 commands · 1 agent
Notify a session's progress to your phone or inbox - a two-way Telegram bot (questions wait for your reply, topic-per-job dispatcher) or email over SMTP/an MCP connector.
1 skill · 1 command
Peer messaging between Claude Code sessions across machines - a self-hosted WebSocket bridge, an MCP client per session, a hook that wakes an idle session on a new message, and busy/idle/waiting state hooks its installer writes into your settings. Hardened fork of Peuqui/AI-Connect (MIT): per-peer tokens, no name takeover, peer content framed as untrusted, VPN-scoped listener. The watcher hook runs once installed and is silent until the plugin's own installer has run; that installer registers the MCP server and the state hooks.
hooks