Claude Ecosystem Research
This file captures official Claude and Anthropic sources that should guide V1. It intentionally avoids leaked/internal material and does not treat third-party claims as product truth.
Skills
Sources:
Key implications for V1:
- A skill is a directory with
SKILL.md as the entrypoint and optional supporting files such as examples, templates, references, and scripts.
- Skills can be invoked directly or loaded when Claude decides they are relevant.
- Project skills, personal skills, enterprise-managed skills, commands, and plugin skills have different scopes and precedence.
- Current Claude Code docs say custom commands have merged into skills conceptually, but existing command files still work.
- Claude Code adds features beyond the open Agent Skills standard, including invocation control, subagent execution, dynamic context injection, path restrictions, and tool restrictions.
description is the key field for skill discovery, and descriptions should lead with the use case.
V1 opportunity:
- Upgrade Skill Quality Doctor to validate official-style fields and explain how each issue affects invocation, sharing, or context cost.
- Add a Template Gallery that teaches the difference between reference skills, workflow skills, command-style skills, and subagent-backed skills.
- Add import preview for skill folders with supporting-file awareness.
Commands
Source:
Key implications for V1:
- Commands control Claude Code sessions and include built-ins, bundled skills, and workflows.
- The commands docs position skills as the way to add custom commands.
- Commands such as
/skills, /reload-skills, /mcp, /agents, /doctor, /code-review, /batch, /run, and /verify show that Claude Code is becoming a broader workflow surface.
V1 opportunity:
- Add read-only inventory for project command files.
- Explain command/skill overlap in the UI.
- Suggest
/reload-skills when a user changes skill content and the active Claude session may need a reload.
Subagents
Source:
Key implications for V1:
- Subagents are Markdown files with YAML frontmatter and a system prompt body.
- Subagents can be personal, project-level, plugin-provided, or managed.
- Subagent definitions include tool restrictions, model choice, permission mode, skills to preload, memory, background behavior, and worktree isolation.
- Project-level subagents are useful for team-shared workflows.
V1 opportunity:
- Add read-only project subagent inventory.
- Add a future template for skill content that is meant to run in a subagent.
- Warn when subagent names appear duplicated or missing required frontmatter.
MCP
Source:
Key implications for V1:
- MCP connects Claude Code to external tools and data sources.
- Project-scoped MCP servers can be shared through
.mcp.json.
- MCP servers have local and remote transports, auth flows, status commands, output limits, and prompt/resource surfaces.
- Claude Code can also expose its own tools as an MCP server.
V1 opportunity:
- Add read-only
.mcp.json inventory and schema sanity checks.
- Show MCP readiness as a local project diagnostic, not as a managed credential viewer.
- Detect high-risk patterns such as broad local command servers and missing project documentation.
Settings And Hooks
Sources:
Key implications for V1:
- Claude Code settings are scoped as managed, user, project, and local.
- Project settings are shared with collaborators; local settings are machine-specific.
- Many settings reload in a running session, while some require restart or a command.
- Hooks are lifecycle automation points and can run commands or external endpoints.
V1 opportunity:
- Add read-only settings and hooks inventory.
- Make Setup Doctor explain which project-level config is shareable and which local config should stay private.
- Add privacy warnings for config files that may contain machine-specific paths or secrets.
Agent SDK
Source:
Key implications for V1:
- The Agent SDK provides the same broad agent loop and tool model that powers Claude Code, but as a library.
- Official docs position CLI for one-off and interactive development, and SDK for production automation.
V1 opportunity:
- Keep V1 on the CLI/API path.
- Document SDK as a future integration track, not a V1 dependency.
- Avoid adding SDK behavior until the local CLI product is polished.
Claude Education
Source:
Key implications for V1:
- Learning Mode emphasizes guiding rather than answering, Socratic questioning, core concepts, templates, and rubrics.
- Education use cases include research, writing, feedback, rubrics, and administrative knowledge conversion.
V1 opportunity:
- Add Guided Skill Builder that asks users to define purpose, audience, trigger examples, inputs, boundaries, and quality criteria before drafting.
- Add rubric feedback for skill drafts instead of one-click generation.
- Add templates and examples that help users learn why a skill is structured a certain way.