Capture how your team works
Store code rules, general guardrails, project initialization recipes, and PR review guidelines with revision history.
Agent policy, owned by humans
Give every agent your team’s better judgment.
Laguarde keeps engineering rules, safety boundaries, project recipes, and review practices in one place—then makes them available exactly when an agent is about to act.
Hand the setup to your agent.
Paste this prompt into an agent working inside the repository. It tells the agent to reuse an existing Laguarde daemon, register the project, install the policy-gate skill locally, configure its own MCP connection, and verify the result.
Install and configure Laguarde for this repository. First read the Laguarde installation contract and follow it exactly: …/install
Before starting anything, check whether a Laguarde daemon is already healthy on this computer and reuse it. Do not create a second instance. Register the current Git repository and configure this agent/client to use the returned project-specific MCP URL.
Install the Laguarde policy-gate skill for this project, not globally:
npx -y skills@latest add https://github.com/FuturPanda/laguarde --skill laguarde-policy-gate --yes
Read and follow the installed laguarde-policy-gate skill. Verify that the Laguarde MCP tools are available, then load the policy bundle for this project. Do not perform other material actions until the project-bound policies have loaded and Laguarde has allowed and recorded the exact action.
When finished, report the project ID, dashboard URL, MCP URL, skill installation scope, and verification result. Do not expose secrets or create or modify policies during setup.A control loop the whole team can understand.
Agents receive context and a precise boundary. Humans retain the decisions that matter, and recurring feedback becomes durable policy instead of disappearing into one chat.
Evaluate before acting
The agent declares its intended files, commands, and dependencies. Laguarde returns a decision tied to exact policy revisions.
Turn feedback into policy
Related corrections converge into a proposal. The agent may suggest; only a human can ratify the next accepted revision.
Four outcomes. No vague “be careful.”
Every evaluated action receives a visible next step, from normal work through explicit human approval to a hard stop.
AllowedProceed inside the declared scope and retain test evidence.
LimitedNarrow, split, or clarify the action before evaluating it again.
ApprovalWait for a human to approve this exact recorded action.
ForbiddenDo not execute. Propose an alternative that avoids the unsafe operation.
Run it where your policies belong.
Run one private local daemon for all of your projects, or deploy one server behind your team’s HTTPS endpoint. Each repository gets its own identity and MCP URL while sharing global policy.
Local MCP with npm
Reuse the local daemon and register the current repository.
npx -y --package laguarde-mcp@0.3.2 laguarde-daemon register --cwd .Team server with Docker
Good for a shared policy registry on a trusted private network.
docker compose up --build
The honest boundary: MCP makes policy discoverable
and decisions auditable. Technical enforcement requires the host
agent, command proxy, sandbox, or CI system to make Laguarde a
mandatory step before execution.