Skip to content

bundled-skills-catalog.md

Latest commit

 

History

History
288 lines (213 loc) · 10.6 KB

bundled-skills-catalog.md

File metadata and controls

288 lines (213 loc) · 10.6 KB
Raw

Bundled Skills Catalog

Generated: 2026-04-01 Extraction basis:

  • bundled skill registry
  • bundled skill definitions

Scope

This document catalogs the skills that ship in the product binary.

It does not cover:

  • user-installed skills
  • marketplace plugin skills
  • built-in plugin skills

It is meant to be self-contained. The descriptions below explain what each skill is for, what kind of workflow it automates, and what makes it materially different from the others.

1. What A Bundled Skill Is

A bundled skill is a shipped workflow prompt that the product already knows how to invoke. It is not:

  • a low-level execution tool
  • a user-installed extension
  • a marketplace plugin

Instead, it is a reusable operating procedure packaged with the product. A skill usually does one of these things:

  • teaches the model a built-in workflow
  • narrows the allowed tools for that workflow
  • changes how the model should interact with the user for that workflow
  • packages domain-specific guidance or documentation for a recurring task

2. Always-Available Bundled Skills

update-config

  • purpose: configure Claude Code via settings.json
  • focuses on hooks, permissions, env vars, and settings troubleshooting
  • allowed tools: Read
  • user-invocable: yes
  • practical meaning: this is the built-in settings expert. It exists to explain how Claude Code is configured and how to change that configuration safely.

keybindings-help

  • purpose: help customize keyboard shortcuts and ~/.claude/keybindings.json
  • allowed tools: Read
  • user-invocable: no
  • enabled only when keybinding customization is enabled
  • practical meaning: this is a hidden helper skill used when the product needs to explain or generate keyboard shortcut configuration.

verify

  • purpose: verify that a change actually works by running or exercising the thing that changed
  • user-invocable: yes
  • bundled reference files: examples/cli.md, examples/server.md
  • practical meaning: this is the product’s built-in “prove it works” workflow. It pushes the model away from static reasoning and toward real validation through app runs, CLI checks, or other concrete verification steps.

debug

  • purpose:
    • ant build: debug the current Claude Code session by reading session debug logs
    • external build: enable debug logging and diagnose issues
  • allowed tools: Read, Grep, Glob
  • disables automatic model invocation
  • user-invocable: yes
  • practical meaning: this is the built-in incident triage workflow for Claude Code itself, not for arbitrary app debugging in general.

lorem-ipsum

  • purpose: generate filler text for long-context testing
  • argument: token count
  • user-invocable: yes
  • ant-only in practice
  • practical meaning: this is a utility skill used to manufacture large amounts of text for harness, prompt-window, or context-behavior testing.

skillify

  • purpose: turn a successful session workflow into a reusable custom skill
  • allowed tools:
    • Read
    • Write
    • Edit
    • Glob
    • Grep
    • AskUserQuestion
    • Bash(mkdir:*)
  • disables automatic model invocation
  • user-invocable: yes
  • practical meaning: skillify is a workflow-capture and packaging assistant. It studies the current session, identifies the repeatable process the user and model just went through, interviews the user about the desired reusable version, drafts a SKILL.md, asks for confirmation, and saves the new skill either to the repo or the user’s personal skills directory.
  • what a reader should understand without source access: skillify is how the product teaches itself new workflows from real usage. It is a built-in skill authoring assistant.

remember

  • purpose: review auto-memory entries and promote or reconcile them across memory layers
  • user-invocable: yes
  • enabled only when auto-memory is enabled
  • practical meaning: this is the memory curation workflow. It lets the model inspect what has been auto-learned and decide what should be kept, merged, or promoted into a more durable memory layer.

simplify

  • purpose: review changed code for reuse, quality, and efficiency, then fix issues found
  • user-invocable: yes
  • practical meaning: this is a built-in self-review and cleanup workflow. It asks the model to critique and tighten its own code changes after implementation.

batch

  • purpose: plan a large-scale change and execute it in parallel across many isolated worktree agents
  • whenToUse: sweeping, decomposable multi-file migrations or refactors
  • disables automatic model invocation
  • user-invocable: yes
  • strong behavioral contract: research first, enter plan mode, split work into 5-30 units, determine an end-to-end verification recipe, then launch isolated background worktree agents that each open a PR
  • practical meaning: this is the product’s built-in bulk-change orchestrator. It is not just “do many edits”; it is “research, decompose, distribute, verify, and track a large migration.”

stuck

  • purpose: investigate frozen or slow Claude Code sessions and post a diagnostic report
  • marked ant-only in description
  • user-invocable: yes
  • operational contract: inspect other Claude Code processes, detect high CPU / hung / zombie / stopped sessions, inspect child processes and debug logs, and post a diagnostic report if a stuck session is found
  • practical meaning: this is an operational support skill for diagnosing unhealthy Claude Code processes on the local machine.

3. Feature-Gated Bundled Skills

dream

Availability: only when dream/background-memory features are enabled.

Interpretation:

  • this is tied to the “dream” or background-memory product surface
  • available evidence indicates it belongs to the background-memory / deferred-work family

hunter

Availability: only when review-artifact features are enabled.

Interpretation:

  • review-artifact / hunter workflow is not always present
  • available evidence indicates it belongs to the deep-review / artifact-review family

loop

Availability: only when recurring-trigger features are enabled.

  • purpose: run a prompt or slash command on a recurring interval
  • whenToUse: recurring polling or repeated tasks, not one-offs
  • user-invocable: yes
  • final visibility also depends on isKairosCronEnabled()
  • practical meaning: loop is the local recurring-work scheduler for prompts and commands.

schedule

Availability: only when remote recurring-agent features are enabled.

  • purpose: create, update, list, or run scheduled remote agents on a cron schedule
  • whenToUse: recurring remote-agent automation
  • allowed tools: REMOTE_TRIGGER_TOOL_NAME, ASK_USER_QUESTION_TOOL_NAME
  • also requires:
    • feature value tengu_surreal_dali
    • policy allowance for remote sessions
  • practical meaning: this is the remote counterpart to loop: instead of repeating a local prompt, it manages recurring remote agents.

claude-api

Availability: only when app-building / Claude API features are enabled.

  • purpose: build apps with the Claude API or Anthropic SDK
  • trigger guidance is embedded in the description
  • allowed tools: Read, Grep, Glob, WebFetch
  • user-invocable: yes
  • packaging detail: this skill bundles a large inline documentation corpus for Python, TypeScript, Java, Go, Ruby, C#, PHP, curl, prompt caching, tool use, batches, files API, streaming, and agent SDK patterns
  • practical meaning: this is a built-in developer advocate skill. It specializes the model for projects that are explicitly using the Claude API or Anthropic SDKs.

claude-in-chrome

Availability: only when Chrome integration is present and auto-enabled.

  • purpose: automate Chrome to interact with pages, forms, screenshots, console logs, and navigation
  • should be invoked before direct mcp__claude-in-chrome__* use
  • allowed tools: Chrome MCP tool set
  • user-invocable: yes
  • visibility depends on shouldAutoEnableClaudeInChrome()
  • activation behavior: the prompt explicitly tells the model to begin with the browser tab context MCP call before using lower-level browser tools
  • practical meaning: this is the browser-automation entry skill. It converts raw browser MCP capability into a guided workflow.

run-skill-generator

Availability: only when internal or experimental skill-generation features are enabled.

Interpretation:

  • there is an internal or experimental skill-generation workflow not always exposed
  • available evidence indicates it is related to automated or semi-automated skill generation

4. Product Signals From The Skill Set

The bundled skills reveal several product bets:

  • settings and harness customization are important enough to ship dedicated guidance (update-config)
  • repeatable workflow capture is a first-class UX (skillify)
  • background / scheduled / remote automation is a real feature family (loop, schedule, dream)
  • browser automation is part of the built-in vision (claude-in-chrome)
  • large-scale parallel change execution is explicitly encouraged (batch)
  • memory curation is becoming productized (remember)

5. Bundled Skill Capability Matrix

The bundled skills divide into a few clear families:

  • configuration and harness help: update-config, keybindings-help, debug
  • verification and improvement: verify, simplify
  • workflow capture and memory: skillify, remember
  • orchestration and automation: batch, loop, schedule, dream
  • browser and API building: claude-in-chrome, claude-api
  • diagnostics and support: stuck
  • utility or testing: lorem-ipsum, hunter, run-skill-generator

This matters because bundled skills are not a random grab bag. They encode the product team’s opinionated workflows.

6. Registration Gates At A Glance

Always registered:

  • update-config
  • keybindings-help
  • verify
  • debug
  • lorem-ipsum
  • skillify
  • remember
  • simplify
  • batch
  • stuck

Conditionally registered:

  • dream when KAIROS or KAIROS_DREAM
  • hunter when REVIEW_ARTIFACT
  • loop when AGENT_TRIGGERS
  • schedule when AGENT_TRIGGERS_REMOTE
  • claude-api when BUILDING_CLAUDE_APPS
  • claude-in-chrome when Chrome integration auto-enables
  • run-skill-generator when RUN_SKILL_GENERATOR

This gate inventory is useful because it distinguishes shipped knowledge that is always present from shipped knowledge that is rollout- or feature-dependent.

7. Relationship To Other Extension Surfaces

Bundled skills are only one extension layer.

The repo also supports:

  • built-in plugins
  • marketplace plugins
  • user skills
  • dynamic skills

But the bundled skills catalog is still valuable because it shows the opinionated workflows the binary ships by default.