crew

Plugin crew in Adminhead/claude-plugins. Commands read from the repo's manifest — today.

Subagents (15)

• Explore

  Fast read-only codebase navigator. Use proactively to locate code, map how a feature works, find callers and definitions, or answer "where/what/how is X" without flooding the main context. Specify breadth — "quick", "medium", or "very thorough". Not for failures with an unknown cause (debugger) or p

• architect

  Architect for implementation planning. Use proactively before multi-slice, cross-cutting, risky, or unclear-approach work, and when choosing between designs. Explores the code, drafts a sliced plan, stress-tests it with an adversarial external critic (codex/agy via review:second-opinion), and return

• browser-tester

  Verifies a running web app in a real browser against acceptance criteria. Use proactively after UI or user-journey changes, when tests alone don't show the feature works. Not for writing test specs (coder with testing:e2e) or fixing failures. Returns a pass, fail, or blocked verdict per criterion wi

• coder-high

  Implementer for a bounded feature, fix, integration, or refactor, typically 1–5 files. Use proactively when the change needs design judgment (structure, interfaces, trade-offs) or coder escalated. Not for work with a clear, known approach (coder), multi-slice or unclear-approach work (architect firs

• coder-max

  Last-resort implementer for work that needs deep reasoning. Use only after coder-high failed with a known cause, or when architect, debugger or debugger-critical marked the work high-complexity (subtle concurrency, tricky algorithms, cross-cutting invariants, a hard fix for a diagnosed bug). Not for

• coder

  Default implementer for a bounded change with a clear, known approach, including mechanical multi-file edits (renames, config values, import updates, applying a specified patch). Use proactively for such work so the main session stays free. Not for work needing design judgment (coder-high), multi-sl

• dba

  Database implementer — schema design, migrations, indexes, query performance, and data integrity constraints. Use proactively for such changes. Never runs migrations or writes against shared, staging, or production databases unless the brief says so. Not for application code (coder, coder-high), CI

• debugger-critical

  Read-only root-cause diagnosis for failures debugger could not confirm, and for high-stakes failures (concurrency, data loss or corruption, security, production incidents). Use proactively when debugger hands back `escalate: debugger-critical`. Not for first-pass diagnosis (debugger), fixes with a k

• debugger

  Read-only root-cause diagnosis for failures with an unknown cause. Use proactively for failing or flaky tests, unexpected behavior, errors with no known cause, and regressions, before anyone edits code. Hands off to debugger-critical when it can't confirm the cause or the failure involves concurrenc

• devops

  Infrastructure and delivery implementer — CI/CD workflows, Dockerfiles and compose, Kubernetes/Helm/GitOps manifests, Terraform/OpenTofu/Terragrunt, Ansible, and deploy and runtime config. Use proactively for such changes. Never applies changes to live infrastructure unless the brief names the actio

• orchestration

  Main-session orchestrator, loaded with `claude --agent crew:orchestration` by the launch script. Plans work, routes it to the other agents, and verifies what they return. Not for spawning as a subagent.

• quick-review

  Fast, focused read-only review of small, task-scoped changes. Use proactively after routine implementations to check the diff against the task's acceptance criteria. Not for large, risky, or security-relevant changes (review). Returns a short pass, or issues marked blocker or minor.

• review

  Independent, thorough read-only review of implementations and plans. Use proactively after large or risky changes (auth, data, concurrency, migrations, public APIs, IaC, CI workflows), and on high-stakes plans before execution. Can bring in codex/agy (review:second-opinion) when a different model fa

• scribe

  Documentation writer and updater — READMEs, guides, docstrings, changelogs, ADR prose, CLAUDE.md/AGENTS.md wording, and keeping docs in sync after code changes. Use proactively when a change leaves docs stale. Not for code changes (coder) or decisions about what the docs should claim beyond what the

• security-auditor

  Read-only security review of a diff or code area. Use proactively after changes to auth, input handling, secrets, crypto, dependencies, CI workflows, or anything crossing a trust boundary, alongside review (which covers correctness). Not for general correctness (review), fixes (coder, devops, dba),

Hooks (6)

• PreToolUse

  bun --no-env-file --config=/dev/null "${CLAUDE_PLUGIN_ROOT}/scripts/readonly-gate.ts" || exit 2

• PreToolUse

  python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/readonly-gate.py" || exit 2

• Stop

  python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/orchestration-guard.py" stop

• UserPromptSubmit

  python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/orchestration-guard.py" prompt

• PostToolUse

  python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/orchestration-guard.py" ledger

• PostToolUse

  python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/orchestration-guard.py" edit