CrewRig
CrewRig

Your team's AI context, built once — not rebuilt by everyone.

CrewRig is a shared configuration layer for AI coding agents. Profile, conventions, skills, and memory live in one repo, deploy to every CLI, and get sharper the more your team uses them.

Fork on GitHub → See how it works ↓

Open source · Works with Claude Code, Gemini CLI & GitHub Copilot CLI

Layered context

The AI that forgets who it's working with

Priya Nair — staff engineer, platform team

Every engineer on Priya's squad starts each AI session from zero. The model doesn't know their stack, their review rituals, or that the team settled on a pattern three quarters ago. Priya ends up re-explaining the same conventions in prompt after prompt — and each teammate explains them slightly differently, so the AI behaves slightly differently for everyone.

CrewRig stacks configuration into priority-ordered layers (00–60): agent identity, seniority, organization policy, personal profile, role expertise, and team norms. Each engineer's profile is personal; the team and expertise layers are shared. The agent loads the full context automatically, so it arrives already knowing how Priya's team works — and behaves consistently for everyone who inherits the same layers.

Priya Nair arranges translucent stacked panels of violet light along a glass office wall while teammates work in soft background focus.

Shared cross-tool memory

The context that walks out the door

Marcus Bell — senior backend engineer

Marcus has spent months teaching his agent the quirks of Quaymont's freight-routing service — the edge cases, the workarounds, the reasons behind odd decisions. But all of it lives in his local session history. When he switches machines, or a teammate picks up the service, that hard-won context is gone, and the next agent starts blind.

CrewRig wires agents into MemPalace, a persistent memory layer that survives across sessions and across tools. What an agent learns in one session — decisions, obstacles, the reasoning behind a fix — is written once and readable later, by Marcus or by a teammate's agent, whether they're on Claude Code, Gemini CLI, or Copilot.

Marcus packs up to leave in the evening while a warm glowing archive of memory drawers persists behind him, its violet light reaching toward Lena's screen nearby.

Skill, agent, and command authoring & sharing

Written once, somehow rewritten three times

Lena Ostrowski — mid-level full-stack engineer

Lena writes a genuinely useful agent skill — a review helper tuned to the team's conventions. A week later she finds Marcus has built almost the same thing for Gemini CLI, because hers only worked in Claude Code. The capability the team needed already existed; it just couldn't travel.

In CrewRig, skills, agents, and commands are authored once as a single Markdown file in artifacts/. One build step (scripts/build-components.sh) compiles that source into outputs for Claude Code, Gemini CLI, and GitHub Copilot CLI. Lena writes the skill one time; her teammates install it on any supported CLI.

Lena watches a single authored document fan out into three identical glowing copies flowing toward terminals labelled Claude Code, Gemini, and Copilot, as Marcus and Aisha lean in.

Harness feedback loop

The papercut that never gets fixed

Tomas Reyes — engineering lead

Tomas watches his squad hit the same small frictions with their AI tooling week after week — a misleading prompt, a tool that does the wrong thing, a workflow step that's gone stale. Everyone grumbles in standup; nobody files it; the rough edge survives forever because reporting it costs more than working around it.

CrewRig builds the feedback loop in. When an agent hits friction during real work, it tags it via the harness-report skill into a shared store. The harness-curator then clusters those tags by theme and opens one GitHub issue per cluster. And because each fix ships back into the shared config, one engineer's papercut becomes everyone's improvement — the whole team's tooling gets sharper from each person's friction, instead of everyone routing around the same wall alone.

Tomas at a standup board where scattered sticky-note frictions converge and resolve into a few clean labelled issue cards, with Priya, Marcus, and Lena gathered around.

Multi-CLI parity

Switch the tool, rebuild everything

Aisha Diallo — DevX / tooling engineer

Aisha moves between Claude Code, Gemini CLI, and Copilot depending on the task. Without a shared layer, each tool is its own island: her profile, her skills, her team's conventions all have to be rebuilt per CLI. Trying a different tool means rewriting her whole setup — so in practice, nobody does.

CrewRig holds one source configuration in config/ and artifacts/, and its setup and build scripts deploy it into each CLI's own directory. The same layered context and the same skills run on Claude Code, Gemini CLI, and GitHub Copilot CLI. Aisha switches tools without rebuilding her setup — the context follows her.

Aisha at a multi-monitor setup with three CLIs open side by side, a single continuous ribbon of violet light bridging the same configuration across all three screens.

Up and running in minutes.

No accounts, no SaaS, no waiting list.

Step 1 — Clone the repo 
git clone https://github.com/crewrig/crewrig.git

Get a local copy of the framework.

Step 2 — Install prerequisites

Read the README → Prerequisites and install the required tools:
Task  ·  Claude Code, Gemini CLI, or GitHub Copilot CLI  ·  fzf  ·  uv  ·  yq

OS-specific install commands are in the README.

Your AI harness:
Step 3 — Initialize 
claude /init-personal-profile
claude /init-soul

Build your personal profile and customize the agent identity.

Step 4 — Setup 
task setup-claude-interactive

Deploys the shared config to your Claude Code harness.

View on GitHub →