Skip to main content
command CX Codex

toolkit — Claude Code command: what it does, how to install

The toolkit command maps project tasks to tools and skills, noting the best option, a fallback and when to skip it; it runs inside /team runs.

MIT ★ 328 updated 2026-10-29

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/commands/toolkit.md
Open on GitHub

What toolkit does

The command opens a registry file that pairs every capability with the tool or skill that handles it best. Entries cover PRD generation, code review, debugging, changelogs. Each one lists the preferred tool, notes on where its output tends to fall short, an alternative, and the situation where you skip it altogether.

During a /team run the orchestrator picks the entries that fit the current action and pastes them into Layer 2 agent prompts. Agents never open the file themselves, so editing it mid-run leaves that run unchanged.

When a run ends, Layer 4 principals can suggest registry updates based on Layer 3 review findings, and the orchestrator applies them. A repo or skill you stumble on can go in as a fresh entry and becomes available to later /team runs. Some entries have no fallback: any claim about tests or coverage has to produce EVIDENCE.md. Standalone, outside /team, the command has nothing to act on.

What it helps with

  • pick the right skill for PRD generation instead of a hardcoded one
  • see up front when a tool is worth skipping
  • register a newly found repo so later runs can use it
  • find the fallback when the preferred skill is unavailable
  • stop a test report from passing without EVIDENCE.md

Example prompts for the agent

  • Open /toolkit and show which tool the registry prefers for code review.
  • Add the changelog skill I just found to /toolkit, with a note on when to skip it.
  • Before the /team run, pull the PRD generation entries from /toolkit along with their quality notes.

How to install toolkit

Manually: put the files in place

.claude/commands/toolkit.md

Call it with /toolkit. For all projects use ~/.claude/commands/

Questions about toolkit

How do I install toolkit in Claude Code?

Place the command file at .claude/commands/toolkit.md in your project; it then runs as /toolkit.

What does toolkit do?

It keeps a registry that matches each capability to a preferred tool or skill, with quality notes, an alternative and a skip rule.

Does it work without a /team run?

No. The orchestrator reads the registry and inlines matching entries into agent prompts, and agents never read the file on their own.

Source text

toolkit.md · MIT
---
description: "Use when a /team run must pick the best tool or skill for a capability, or when adding a discovered tool to the registry. Maps each capability to its best tool, quality notes, fallback, and when-to-skip rule."
---

# Team Toolkit Registry

> **Purpose:** Maps capabilities to the best available tool/skill.
> Layer 2 agents consult this before doing work — they use the best tool
> available, not a hardcoded one.
>
> **How it's used:** The orchestrator (team.md) reads this file, extracts
> entries relevant to the current action, and inlines them into Layer 2
> agent prompts. Agents never read this file directly.
>
> **Updating:** After every /team run, Layer 4 principals may recommend
> toolkit updates based on Layer 3 review findings. The orchestrator
> applies these updates.
>
> **Adding new tools:** When you discover a new repo, skill, or tool,
> add an entry here. It becomes available to all future /team runs.

---

## PRD Generation

- **Best tool:** prd-generator (skill; the PM Studio suite refers to it as /pmstudio-prd, but the skill declares itself as `prd-generator` and that is the name that resolves)
- **Quality notes:**
  - Always include a rollback plan in the launch section
  - Make success metrics measurable (not "improve UX" — use quantitative targets)
  - NFR section tends to be generic — add project-specific NFRs
- **Alternative:** Direct writing by senior-pm role
- **When to skip tool:** PRD < 3 sections or highly specialized format

## Stakeholder Communications

- **Best tool:** /pmstudio-comms
- **Quality notes:**
  - Subject lines sometimes too formal — shorten and make actionable
  - Add TL;DR for emails > 3 paragraphs
- **Alternative:** Direct writing by comms-specialist role
- **When to skip tool:** Simple 1-paragraph updates

## Architecture Review Decks

- **Best tool:** /pmstudio-arb
- **Quality notes:**
  - 11-slide consulting format is solid
  - Data visualization sections need improvement — pair with data-viz-specialist
  - This entry is a presentation generator, not an architecture tool. It formats an architecture for a review board audience; it does not analyse or decide one. For analysis use the Architecture Audit entry, and for the underlying structural facts use the Codebase Mapping and Code Graph Queries entries.
- **Alternative:** Direct creation by structured-presentation + narrative-architect roles
- **When to skip tool:** Non-consulting format presentations

## Meeting Notes

- **Best tool:** /pmstudio-meeting-notes
- **Quality notes:** No findings yet
- **Alternative:** Direct writing by business-analyst role
- **When to skip tool:** Quick standup notes (< 5 bullet points)

## Change Logs

- **Best tool:** /pmstudio-changelog
- **Quality notes:** No findings yet
- **Alternative:** Direct writing from git log
- **When to skip tool:** Single-item changes

## Disaster Recovery Plans

- **Best tool:** /pmstudio-dr
- **Quality notes:** No findings yet
- **Alternative:** Direct writing by senior-cloud-architect + sre-devops roles
- **When to skip tool:** Non-SaaS products

## Incident Response Plans

- **Best tool:** /pmstudio-irp
- **Quality notes:** No findings yet
- **Alternative:** Direct writing by sre-devops role
- **When to skip tool:** Internal tools with no SLA requirements

## Recovery Runbooks

- **Best tool:** /pmstudio-recovery
- **Quality notes:** No findings yet
- **Alternative:** Direct writing by sre-devops role
- **When to skip tool:** When DR plan doesn't exist yet

## Code Implementation

- **Best tool:** Superpowers pipeline (brainstorming → writing-plans → executing-plans)
- **Quality notes:**
  - TDD enforcement is strong
  - Plan granularity could be finer for complex features
- **Alternative:** Direct coding by engineering roles
- **When to skip tool:** Single-file changes or < 30 lines of code

## Project Orchestration (Multi-Phase)

- **Best tool:** GSD (/gsd:plan-phase → /gsd:execute-phase → /gsd:verify-work)
- **Quality notes:**
  - Good for multi-phase projects with persistent state
  - Overkill for single-session tasks
- **Alternative:** Manual phasing with /team plan
- **When to skip tool:** Single-session scope or < 3 phases

## Test Evidence Protocol

- **Best tool:** `team:evidence.md` (the canonical Test Evidence Protocol)
- **Quality notes:**
  - Any quality claim (tests pass, lint clean, coverage N%) MUST follow this protocol and produce `EVIDENCE.md`. Evidence or it didn't happen.
  - Determine the authoritative gate from CI config (`.github/workflows`, `Makefile`); use the CI-pinned tool versions, never a weaker local equivalent.
  - Skipped tests are `UNVERIFIED`, never a pass. Provision integration deps (e.g. Postgres + DSN) so gated tests actually run, or label the gap.
  - Coverage must be measured (`--cov --cov-branch`) and captured, not narrated.
- **Alternative:** None — this is mandatory for any pipeline that reports test/lint/coverage results.
- **When to skip tool:** Never for code that ships via PR. (Pure-docs runs have no test gate.)

## Test-Driven Development

- **Best tool:** test-driven-development
- **Quality notes:** No findings yet
- **Alternative:** Direct test writing by qa-test-architect role
- **When to skip tool:** When adding tests to existing test suite (just follow patterns)

## Systematic Debugging

- **Best tool:** systematic-debugging
- **Quality notes:** No findings yet
- **Alternative:** Direct debugging by relevant engineering role
- **When to skip tool:** Obvious bugs (typos, wrong variable names)

## Code Review

- **Best tool:** requesting-code-review
- **Quality notes:** No findings yet
- **Alternative:** /team review action (uses full 4-layer pipeline)
- **When to skip tool:** Quick sanity checks on < 50 lines

## Document Sync

- **Best tool:** /pmstudio-sync
- **Quality notes:** No findings yet
- **Alternative:** Manual sync
- **When to skip tool:** Single-document updates

## NFR Audit

- **Best tool:** /pmstudio-nfr
- **Quality notes:** No findings yet
- **Alternative:** Manual checklist review
- **When to skip tool:** Early-stage projects (not enough artifacts to audit)

## Confluence Publishing

- **Best tool:** Direct curl REST API calls (v2)
- **Quality notes:**
  - MCP tools don't reliably load as deferred tools in Claude Code
  - Always fetch current version before updating (version increment required)
  - Images use ac:image storage format — upload as attachments first
- **Alternative:** Manual publishing
- **When to skip tool:** N/A — curl is the reliable path

## Architecture Conformance

- **Best tool:** `team:architecture.md` (the canonical Architecture Conformance Protocol), enforced by `~/.claude/skills/arch-index/scripts/validate_index.py`
- **Quality notes:**
  - Paths or it does not exist. The validator exit code is the gate, and a validator that cannot be executed is a BLOCK rather than a pass. There is deliberately no prose fallback, because a fallback an agent can narrate is the failure this gate exists to remove.
  - Structural drift only. A component whose datastore was swapped inside its own already-claimed directory produces no violation, so never report a clean drift result as evidence that the architecture is sound.
  - An index whose pinned commit is behind HEAD is `STALE`, and every conformance claim resting on it is `UNVERIFIED` until reconciled. Check currency before quoting the index.
  - The index sits at C4-Container altitude. Do not derive per-agent file ownership from it; a single component can legitimately own more than a thousand files.
- **Alternative:** None — this is mandatory for any action that consumes or produces `.arch/index.json`.
- **When to skip tool:** Never for an action that makes an architectural claim. Actions that make no such claim have no architecture gate.

## Codebase Mapping

- **Best tool:** gsd-map-codebase
- **Quality notes:**
  - Spawns four parallel mapper agents that write seven structured documents to `.planning/codebase/`: `STACK.md`, `INTEGRATIONS.md`, `ARCHITECTURE.md`, `STRUCTURE.md`, `CONVENTIONS.md`, `TESTING.md`, and `CONCERNS.md`. The orchestrator receives only confirmations, which keeps context usage low.
  - Prefer reading an existing `.planning/codebase/` map over re-deriving codebase structure inside an agent prompt. Re-deriving it wastes the tokens this tool exists to save.
- **Alternative:** technical-analyst reading the dependency manifest and directory tree directly
- **When to skip tool:** The repository already has a current `.planning/codebase/` map

## Code Graph Queries

- **Best tool:** gitnexus MCP (`context`, `impact`, `api_impact`, `detect_changes`, `route_map`, `tool_map`, `shape_check`, `query`, `cypher`)
- **Quality notes:**
  - A graph that knows real call edges cannot hallucinate a file path, which makes it strictly more reliable than any model-authored description of how modules relate.
  - Use `impact` and `api_impact` before changing a shared interface, and `detect_changes` to establish what actually moved between two commits.
- **Alternative:** `git diff --name-status` plus Grep for call sites
- **When to skip tool:** The repository is not indexed by gitnexus

## Architecture Diagrams

- **Best tool:** c4-architecture
- **Quality notes:**
  - Tabulates the five C4 levels with the audience and the when-to-create rule for each, so it prevents the common error of drawing a component diagram for a stakeholder who needed a context diagram.
  - Render the resulting Mermaid through `/design-mermaid` (beautiful-mermaid), never through the standard Mermaid CDN.
- **Alternative:** Direct Mermaid authoring by an engineering role
- **When to skip tool:** A current diagram already exists for the level being asked about

## Interface Contract Design

- **Best tool:** api-design-principles
- **Quality notes:** No findings yet
- **Alternative:** Direct design by senior-backend-eng
- **When to skip tool:** The change introduces no new or modified public interface

## Architecture Audit

- **Best tool:** /util:architecture-review
- **Quality notes:**
  - Offers a six-part analysis framework with `--modules`, `--patterns`, `--dependencies`, and `--security` scopes, so pass the scope that matches the question rather than running the full audit by default.
- **Alternative:** /team review action (full 4-layer pipeline)
- **When to skip tool:** A full architectural audit is out of scope for the run

## Architecture Doc Suite

- **Best tool:** /util:create-architecture-documentation
- **Quality notes:**
  - Produces C4 model, arc42, ADR, PlantUML, and Structurizr output. Prefer this over hand-authoring whenever the deliverable is a formal published architecture document.
- **Alternative:** /team document action with technical-writer in Layer 2
- **When to skip tool:** No formal published document is required

---

## Adding New Tools

When you discover a new tool, repo, or skill, add an entry following this template:

```
## [Capability Name]

- **Best tool:** [skill name or command]
- **Quality notes:** [empty until first review]
- **Alternative:** [fallback approach]
- **When to skip tool:** [when direct work is faster]
```

© coco-research · MIT · coco-research/coco

Similar slash commands

All slash commands for Agent setup: memory, context, orchestration →

From the same set82

The whole set →
anchored-layout-expand The anchored-layout-expand rule describes one-axis panel expansion with the following content moving in step. Use it for dropdowns, input cards, or panes that push a neighbor. CX Codex +1 ★ 328repo stars chart-scrub-readout chart-scrub-readout — a Claude rule that turns a chart already on screen into something you can scrub: a tracking line, a marker on the polyline, and a tooltip reading out date and value. CX Codex +1 ★ 328repo stars multi-cursor-choreography multi-cursor-choreography is a Claude Code rule for staging several labeled cursors on one canvas. It choreographs team-like motion without camera tracking. CX Codex +1 ★ 328repo stars architecture A slash command that pins down what counts as following the architecture in a /team run and makes every such claim carry a validator exit code. CX Codex +1 ★ 328repo stars arch-index Builds a validated map of repository components in .arch/index.json and checks it for drift against the code. Use it to trace a component to its files or review structural changes after a refactor. CX Codex +1 ★ 328repo stars hyperframes-keyframes hyperframes-keyframes keeps HyperFrames motion seekable: builds keyframes for GSAP, CSS, Anime.js and WAAPI, then verifies frames before a render. CX Codex +1 ★ 328repo stars pmstudio-meeting-notes pmstudio-meeting-notes converts call transcripts into structured Markdown meeting notes. Cross-references people against stakeholder tables, logs decisions, and flags PRD updates. CX Codex +1 ★ 328repo stars hyperframes-cli The hyperframes-cli skill guides HyperFrames projects from composition checks to preview and rendering. It also helps diagnose build and render failures. CX Codex +1 ★ 328repo stars

Project sponsors

Companies that keep this analytics open