local-first capability composition

Stop loading
every tool.

Index skills, prompts, agents, and MCP servers already on your machine. Compose a complementary set, see what each resource adds, and keep gaps visible.

Try the composition lab ↓View source ↗
coverage-aware compose
$ task2tool compose "review API security and release notes"

1. Pull Request Risk Reviewer
  adds review · api · security

2. Release Notes
  adds release · notes

100% lexical coverage · 2 of 8 loaded · local only
Interactive sample

Compose a capability shelf.

This browser demo uses a small illustrative catalog. The CLI applies the same explainable marginal-coverage strategy to files on your machine.

—marginal lexical coverage · deterministic
——Normalized word overlap, not a semantic capability guarantee.
8resources indexed locally
—complementary resources selected
Small surface, useful defaults

Built for the agent layer.

01

Scan what you have

Discover SKILL.md, agent and prompt files, MCP configurations, and portable JSON catalogs recursively.

02

Compose complements

Greedy marginal coverage chooses resources that add new task terms, then discloses every remaining gap.

03

Share without a service

Export deterministic JSON, readable Markdown, or a self-contained searchable HTML report. Zero runtime dependencies.

Your context window is a workspace, not a warehouse.
No cloud account, hosted index, telemetry, or API key
MCP arguments and environment values stay out of reports
Symlinks and generated dependency directories are skipped
Strict TypeScript and Node 20 with no runtime packages