claude-tools

Run quality checks on a task's implementation before it closes to done. Triggers: 'is this ready?', 'run the review gate', 'check my work', 'I think this is done'. Reviews code + spec adherence, runs tests/build. Design review uses taskmaster:spec-review.

Log a lightweight idea into .taskmaster/ideas/. Triggers: '/add-idea', 'save this as an idea', 'remember this idea', 'log this idea', or something worth keeping, not yet a task. Only way - not backlog_idea_create directly.

Log/update/close project bugs in .taskmaster/bugs/. Triggers: 'log a bug', 'this is a bug', 'track this defect', 'I found a bug', 'shelve this for later', 'list open bugs', 'promote B-XX', 'close B-XX'. Also for ambiguous 'issue' phrasing lacking recurring/systemic/outstanding evidence - falls back here. Only correct way to transition a project Bug.

Scan the codebase for TODO/FIXME/HACK/XXX and cross-reference the backlog. Triggers: 'check todos', 'are my todos tracked', 'scan for todos', 'todo audit', 'what's untracked'.

Write/resolve/drop project decisions in .taskmaster/decisions/. Invoke when about to write an inline option menu - route here, not Options: in chat. Also 'choose between', 'pick an option', 'decide on', 'open question', 'resolve DEC-X', 'drop DEC-X'. Only correct way - do not call backlog_decision_create directly.

Close out a work session. Triggers: 'end session', 'I'm done for today', 'let's wrap up', 'mark this task done', 'save progress'. ONLY correct way to mark tasks done/in-review with a session record.

Write a session handover into .taskmaster/handovers/. Triggers: 'write a handover', 'wrap up', 'for tomorrow', 'before compaction', 'context handoff', 'continue where we left off'. Only way - not backlog_handover_create directly.

Set up Taskmaster in the current project. Triggers: 'set up taskmaster', 'initialize backlog', 'I want to track my work here', or backlog.yaml missing. Offers clean init or project analysis.

Log/update/close project issues in .taskmaster/issues/ - elevated tier for recurring, systemic, or P0/P1 outstanding defects. Triggers: 'log an issue', 'this is an issue', 'file an issue', 'promote to issue', 'is this an issue', 'list open issues', 'triage issues', 'mark issue fixed', 'close ISS-XX'. NOT for one-off bugs, route to taskmaster:bug. Colloquial 'issue' lacking that evidence routes to taskmaster:bug. Only correct way to transition a project Issue.

Manage Taskmaster's Linear sync. Triggers: 'set up linear sync', 'connect to linear', 'link/unlink task to linear', 'show linear status', 'retry linear pushes', 'list linear trackers'. Only way to drive backlog_linear - direct calls skip bootstrap.

Migrate legacy .taskmaster/lessons/ into assistant memory and repo instruction files after a taskmaster 4.x upgrade. Triggers: 'migrate lessons', 'convert lessons to memory', 'what happened to lessons', or start-session finding L-*.md files.

Guided v2 to v3 backlog migration. Triggers: 'upgrade to v3', 'migrate to v3', 'switch to v3', 'enable handovers', 'enable narrative continuity'. Only correct way - do not call backlog_migrate_v3 directly without the pre-flight gate.

Select a task to work on. Triggers: 'pick a task', 'let's work on X', 'start task auth-003', 'what should I tackle next', 'continue this task', 'continue where we left off', a task ID, 'pick this bundle', 'cluster these tasks'. Sets status in-progress, checks deps, creates worktree.

Review a task's implementation plan before writing tests. Triggers: 'review this plan', 'plan review', 'challenge my plan', 'is this plan solid'. Full-lane gate between PLAN and WRITE_TESTS; post-implementation review is taskmaster:review-gate.

Pre-implementation design review of a task's spec. Triggers: 'review this spec', 'challenge this design', 'is this the right approach?', 'spec review', 'design review'. Records the spec-review (full) or design-review (standard) gate. Code review uses taskmaster:review-gate.

Start a work session, orient a new conversation. Triggers: 'let's get started', 'what should I work on', 'show me the backlog', 'orient me', or a new conversation where backlog.yaml exists. Shows dashboard, last session, next tasks.

Universal taskmaster work router for projects with backlog.yaml. Invoke for features, bug fixes, tests, refactors, epic planning, or narrative-continuity work (handovers, issues). Skip only for: pure git commits, dedicated PR security reviews, and small direct changes asked for outright (one sitting, no open design decisions, ~<=3 files).

This skill should be used when the user says "join agent relay", "connect this chat", "ask another agent", "send this to Claude", "send this to Codex", "adversarially review my plan", "check the relay inbox", "talk to another session", or requests a direct or group conversation between AI coding sessions.

Pain-point → workflow codifier. Scans a project's session history for recurring friction patterns and emits draft workflow proposals — skill stubs, command outlines, or settings tweaks — ready to become real plugin components. Always uses an adversarial reviewer to filter one-off complaints from genuine repeating pain. Invoke when the user says 'harvest pain points', 'what should we automate', 'find workflows to codify', 'this keeps happening — make it a skill', 'this keeps coming up', 'what are we doing manually that should be automatic', 'what keeps going wrong', 'what should I turn into a skill', 'look at my session history for patterns', 'review my workflow for friction'. Accepts --project <path>; defaults to cwd. Output is advisory — no skill files are created automatically, only proposals routed through taskmaster:add-idea.

Self-audit of claude-tools plugins. Operates on plugin source files (not project artifacts). Finds dead skills (never invoked across transcripts), oversized skill descriptions (token cost), command pruning candidates, and recurring transcript patterns that should become new skills. Always uses an adversarial reviewer. Invoke when the user says 'audit the plugins', 'find dead skills', 'what's eating our tokens', 'prune commands', 'claude-tools self-audit', 'plugin health check', 'which skills are we actually using', 'are our skills still healthy', or after major plugin changes. Output is advisory — routes proposals through taskmaster:add-idea in claude-tools' own backlog.

Time-windowed retrospective on a project's past work. Reads taskmaster artifacts (handovers, lessons, issues, ideas) and git log to produce a markdown report plus concrete proposals routed back through taskmaster. Invoke when the user says 'retro the last week', 'run a retrospective', 'do a post-mortem', 'what did we accomplish this week', 'wrap up the sprint', 'how did we work in CodeMaestro', 'review our recent sessions', 'what worked and what didn't this sprint', 'look back at this project'. Accepts depth=shallow|standard (deep is v1.1) and --project <path> (defaults to cwd).

Install guard hooks into user-scoped ~/.claude/settings.json. Use when the user says "install guard hooks", "set up guard hooks", or "enable guard hooks".

Drop a feedback message into the claude-tools feedback inbox. Use when the user explicitly requests it ('/inbox-feedback', 'log this as feedback', 'send this to claude-tools', 'report this back', 'feedback to the toolmaker') AND proactively, on your own initiative, when you observe friction with a claude-tools-shipped component during a session — a taskmaster skill that errored, a guard-hook that blocked the wrong command, an MCP tool that returned a confusing shape, a skill description that didn't match its behaviour, a slash command that did the wrong thing. Do not fire for generic project friction or external-tool friction; only for claude-tools components (taskmaster, reflect-auto-improve, guard-hooks, statusline, feedback-inbox itself, ue5-materials, etc.). Silent on success — no user confirmation needed; the user reviews during /inbox triage.

List and walk pending messages in the claude-tools feedback inbox, routing accepted items to taskmaster:add-idea or taskmaster:bug and archiving the rest. Invoke when the user says '/inbox', 'triage the inbox', 'process feedback', 'what's in the inbox', 'any feedback to look at', or sits down in the claude-tools repo and asks what's new. Reads messages written by the feedback-inbox:feedback producer. Non-destructive — processed messages move to inbox/processed/<year>/, never deleted.

Use BEFORE running `git worktree remove`, `git worktree prune`, or any worktree cleanup on Windows or against worktrees that contain submodules or have long paths. Routes through the safe deinit-first procedure to prevent the dotfile-cascade incident where `--force` removal wipes top-level `.git/`, `.gitignore`, `.gitmodules`, `.env*`, `.github/`, `.vscode/`, `.ai/design/` in the PARENT checkout. Invoke when the user says "remove this worktree", "clean up worktrees", "delete the worktree", "prune worktrees", or when guard-hooks blocks a worktree-remove command.

Universal entry point for reflecting on past work. Invoke when the user says 'reflect on', 'do a retro', 'let's audit our work', 'how have we been working', 'what should we automate', 'find dead skills', 'review last week', 'pain-point harvest', 'how are we doing', 'process check-in', 'are we improving', 'check our process', 'team health check', or otherwise asks to look back and improve. Routes to the correct lens skill (retro / harvest / plugin-audit) based on intent.

Configure Claude Code to use the gruku-tools custom statusline (distinct from the native /statusline command which opens settings). Use when the user says "set up the custom statusline", "install the gruku statusline", "/statusline:custom-statusline-install", or right after installing the plugin. Also handles uninstall ("remove the statusline", "/statusline:custom-statusline-uninstall").

Generate a new image from a text prompt using Gemini or OpenAI.

Generate, edit, or refine images using Gemini or OpenAI. Auto-detects intent from context.

Route any UE5 / Unreal Engine material, shader, or material-graph authoring request to the correct ue5-materials sub-skill. Invoke this whenever the user asks to build, edit, tweak, or debug a UE5 material, shader effect, or Material Editor graph — even if they don't explicitly name the skill or say "shader". Covers both live editing via MCPBridge and clipboard-paste T3D workflows.

Author UE5 materials from a plain-text description or a known YAML spec. Produces a YAML file that `apply_material.py` applies to Unreal — live via MCPBridge when the Editor is running, or as T3D clipboard text when it isn't. Dispatches HLSL Architect (Opus) and Node Generator (Sonnet) agents for complex designs; falls through to a direct-YAML path for simple, known node setups. Invoke whenever the user asks to build, change, or iterate on a UE5 material, material function, or shader effect.

Look up UE5 material-graph node types in node_registry.yaml and add missing ones. Invoke before generating YAML when you hit an unfamiliar expression class (e.g. ComponentMask, Fresnel, CustomRotator) or when the user asks whether a node or Material Function is already registered. This is the authoritative catalogue of node shapes the clipboard-mode generator can emit.

Apply the Reality Reprojection design system — auto-detects whether to generate new components, convert existing code, or review compliance. Single entry point that routes to the correct sub-skill.

Convert existing HTML/CSS to use Reality Reprojection design system tokens, classes, and patterns

Generate new HTML/CSS components following the Reality Reprojection design system from natural language descriptions

Audit HTML/CSS for Reality Reprojection design system compliance — reports violations with severity and suggested fixes

Bootstrap the Reality Reprojection design system into a target project — copies CSS bundle, adds font imports, sets polarity

Convert existing HTML/CSS to use Reality Reprojection design system tokens, classes, and patterns

Generate new HTML/CSS components following the Reality Reprojection design system from natural language descriptions

Audit HTML/CSS for Reality Reprojection design system compliance — reports violations with severity and suggested fixes