§01 — Overview Sheet 1 / 6

A family of open tools

Open tooling
for machines
that write code.

comind is a team building a growing line of open-source AI tooling. Three tools are out today — a process layer on top of Claude Code, a self-hosted MCP gateway, and a terminal multiplexer for AI agents — with more in active development. All open, all yours to run.

Catalog 3 shipped · +more
Shipped 01
comind-skills
Shipped 02
comind-mcp
Shipped 03
comind-dock
Next
in development
License
MIT / Apache-2.0
Built by
comind.pro
Status
Open source

◆ Independent, composable, self-hostable. Use one, use both. No accounts required to read the source.

├ a growing family · one philosophy ┤
§02 — Product 01 / comind-skills layer on Claude Code

Engineering process layer

23 skills that
ride on top of
Claude Code.

A self-contained workspace — not a server. Drop it next to your project and the agent gains structured workflows: steps, verification gates, and anti-rationalization tables that stop it skipping the hard parts. Skills auto-activate by context, or you invoke them by hand.

define /spec plan /plan build /build verify /test review /review ship /ship
↗ comind-skills on GitHub
comind-skills / .claude / skills 23 entries

comind-skills/

├─ define/ /spec
├ 01 interview-me
├ 02 idea-refine
└ 03 spec-driven-development
├─ plan/ /plan
└ 04 planning-and-task-breakdown
├─ build/ /build
├ 05 incremental-implementation
├ 06 test-driven-development
├ 07 context-engineering
├ 08 source-driven-development
├ 09 doubt-driven-development
├ 10 frontend-ui-engineering
└ 11 api-and-interface-design
├─ verify/ /test
├ 12 browser-testing-with-devtools
└ 13 debugging-and-error-recovery
├─ review/ /review
├ 14 code-review-and-quality
├ 15 code-simplification
├ 16 security-and-hardening
└ 17 performance-optimization
└─ ship/ /ship
├ 18 git-workflow-and-versioning
├ 19 ci-cd-and-automation
├ 20 deprecation-and-migration
├ 21 documentation-and-adrs
└ 22 shipping-and-launch

Usage — clone · bootstrap · build

# 1 · use as a template
git clone github.com/comind-pro/comind-skills my-app

# 2 · tailor it to your project (≤4 questions)
> /init-project "a SaaS that does X for Y"

# 3 · drive the lifecycle
> /spec → /plan → /build → /review → /ship
§03 — Product 02 / comind-mcp

MCP gateway · self-hosted

One gateway for
all your MCP tools.

Agents shouldn't juggle a dozen MCP servers and APIs. comind-mcp aggregates them, lets you curate & combine the tools, and exposes clean virtual endpoints your agents call — with per-agent keys, a secrets vault, schedules and full logs. Self-hosted, zero-infra.

Type
Self-hosted gateway
Aggregates
MCP · HTTP · OpenAPI
Exposes
Virtual MCP (V-MCP)
Auth
Per-agent keys
Extras
Vault · schedules · logs
Stack
Single binary + Postgres
FIG.3 — sources → V-MCP → agents signal flow

Sources

github-mcp12 tools
stripe (OpenAPI)8 tools
internal-crm5 tools

+ add MCP / HTTP / OpenAPI

↓ curate & combine

V-MCP · support-bot

[x] github.create_issue

[x] stripe.refund_charge

[x] crm.lookup_customer

endpoint · …/g/support-bot

↓ scoped keys

Agents

claude-supportag_live_•••3f9a
ops-runnerag_live_•••b71c
Pipeline — connect · curate · build · grant · observe 5 stages
  1. 01

    Connect sources

    Add any MCP server or HTTP/OpenAPI API. The gateway introspects it and lists every tool it exposes.

  2. 02

    Curate tools

    Pick the tools that matter, rename them, tweak schemas, or merge several calls into one composite tool.

  3. 03

    Build a V-MCP

    Bundle curated tools into a virtual MCP — a single clean endpoint that hides the messy upstreams.

  4. 04

    Grant agents

    Issue per-agent keys scoped to specific V-MCPs. Rotate or revoke any key instantly.

  5. 05

    Observe

    Log every call: which agent, which tool, latency, errors. Schedule jobs, store secrets in the vault.

§04 — Lab / what's next family is growing

Several products · more coming

We ship one,
then hunt for
the next problem.

comind is not a single product — it's a growing line of open-source tools for AI coding agents. Three are out today; more are in active development and research. Everything lands on the org first, open from day one.

↗ Watch the org for releases
comind-pro / status 3 shipped · 1+ planned

shipped/

├ comind-skills engineering process layer v1
├ comind-mcp self-hosted MCP gateway v1
└ comind-dock agent terminal multiplexer phase 1

next/

└ •••••••• exploring — problem first research

◆ Placeholders are intentional — names ship when the tools do. Got a problem worth an open tool? Open an issue on the org.

§05 — Philosophy what the family shares

One philosophy across the whole family.

Skills or gateway, the rules are the same: open source, self-hostable, no lock-in, and built so the agent does the disciplined thing by default — not the shortest one.

N.1

Process, not prose

Skills are workflows agents follow — steps, checkpoints, exit criteria — not docs they skim and forget.

N.2

Anti-rationalization

Every skill ships a table of the excuses agents use to skip steps ("I'll add tests later") with the counter-arguments.

N.3

Verification is non-negotiable

Each skill ends with evidence requirements — tests passing, build output, runtime data. "Seems right" never counts.

N.4

Progressive disclosure

The SKILL.md is the entry point; references load only when needed, keeping token usage minimal.

N.5

Battle-tested judgment

Hyrum's Law, the test pyramid, change sizing, Chesterton's Fence, trunk-based dev — Google-grade practice, baked in.

N.6

Open by default

Everything is MIT licensed. Read it, fork it, ship it in your own teams and tools. No lock-in.

§06 — Get started end of sheet

The whole family lives on GitHub

Everything open,
in one organization.

Start at the comind-pro org — every product, the source, the issues, the releases, and whatever ships next. Read it, fork it, self-host it. MIT, no lock-in, no account to look.