crew-chief

Plugin crew-chief from marketplace crew-chief in aykyusuf/crew-chief. Commands read from the repo's manifest — 2 days ago.

Skills (4)

• crew-handoff

  Sets up session handoff files (STATUS.md, tasks.json, PROGRESS.md and a start/end routine in CLAUDE.md) so a new Claude Code session picks up where the last one stopped. Use when the user accepts crew-chief's handoff offer, or asks to set up handoff, progress, or status files for this project. Never

• crew-mode

  Switches crew-chief's delegation mode for the rest of the session: auto (route by difficulty), solo (main session does everything, no subagents), or delegate (hand every independent piece to a tier).

• crew-routing

  Decides who should do a piece of work and how: the main session or a tiered subagent (Haiku, Sonnet, Opus) at which effort, and which orchestration mechanism fits (subagent, fork, background session, /loop, Monitor, dynamic workflow, agent team, routine). Use before delegating or spawning agents, be

• crew-setup

  Installs crew-chief's tiered agents, read-only guard, and routing policy into the current project's .claude/ directory and CLAUDE.md, for use without the plugin (for example after npx skills add). Optionally adds session handoff files.

Subagents (9)

• deep-reader

  Read-only code analyst (Sonnet, medium effort). Use when the answer requires reading several files together: tracing a flow end to end, finding the root cause of a bug from code and logs, mapping the impact of a planned change, summarising an unfamiliar module, or reading the code together with curr

• implementer-hard

  Implements hard code changes end to end (Opus, high effort). Use for scoped but difficult work that can run without back-and-forth with the user: changes spanning several modules, algorithms, geometry, concurrency, debugging with an unknown root cause, or a task the implementer agent stopped on. Do

• implementer

  Implements well-specified code changes (Sonnet, medium effort). Use when the task names the files, the change, and how to verify it: adding a parameter, writing a test, fixing a located bug. Rote edits that need no judgement (renames, the same line in many files) go to mechanic. Do not use when requ

• mechanic

  Cheap mechanical edits with a stated check (Haiku, low effort). Use for renames, rote find-and-replace across named files, applying a diff the caller already wrote, bumping versions, adding the same line to many files. The brief must list the exact files and a command that verifies the result. Do no

• planner

  Read-only planner (Opus, high effort). Use before implementer-hard on a change that spans several modules or has real design choices: it returns the files to touch, the steps in order, the risks, and a testable done condition, so the build can start in a fresh context. Never edits project files. Do

• reviewer

  Independent read-only reviewer (Opus, high effort). Use before a commit or pull request of a significant or risky change, when the user asks for a review, or when a claim or test suite needs an independent check. Not after every small edit. Because it does not share the author's context, it catches

• scanner

  Cheap, fast read-only lookup (Haiku, low effort). Use proactively for mechanical searches: where a symbol is defined, which files use X, listing files, or running a command and returning only the failing lines from long log or test output. Do not use when the answer needs interpretation, root-cause

• ui-smoke

  Fast, read-only smoke check of a running UI (Haiku, low effort): is the page or screen up, does it show text X, any console or logcat errors. Uses page text, DOM, and logs, not screenshots. Use for a quick yes/no after a deploy or rebuild. Do not use for multi-step flows, clicking through the app, o

• ui-tester

  Runs multi-step UI checks in a browser, Android emulator, or iOS simulator and reports pass/fail with evidence (Sonnet, medium effort). Use for flows, regression passes, or anything needing more than a handful of screenshots, so the images stay out of the main context. Do not use for a single quick

Hooks (12)

• SessionStart

  "${CLAUDE_PLUGIN_ROOT}/scripts/session-policy.sh"

• SessionStart

  "${CLAUDE_PLUGIN_ROOT}/scripts/compact-restore.sh"

• SessionStart

  "${CLAUDE_PLUGIN_ROOT}/scripts/install-row.sh"

• SessionStart

  "${CLAUDE_PLUGIN_ROOT}/scripts/skill-offer.sh"

• UserPromptSubmit

  "${CLAUDE_PLUGIN_ROOT}/scripts/mode-guard.sh" record

• UserPromptSubmit

  "${CLAUDE_PLUGIN_ROOT}/scripts/plugin-notices.sh" prompt

• UserPromptSubmit

  "${CLAUDE_PLUGIN_ROOT}/scripts/skill-hint.sh"

• UserPromptExpansion

  "${CLAUDE_PLUGIN_ROOT}/scripts/mode-guard.sh" record

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/scripts/readonly-guard.sh"

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/scripts/mode-guard.sh" check

• PostToolUse

  "${CLAUDE_PLUGIN_ROOT}/scripts/progress-hook.sh"

• SubagentStart

  "${CLAUDE_PLUGIN_ROOT}/scripts/subagent-start.sh"

Similar plugins