Skillscript CLI
A language and runtime for skills an agent writes and a human approves: execution through allowlisted connectors and signatures
- 76 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Scheduled skills run commands and write files once the operator allowlists them
- Approving a skill opens connectors that hold working credentials
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/skillscriptDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Scheduled skills run commands and write files once the operator allowlists them
- Approving a skill opens connectors that hold working credentials
This is third-party code. Review the repository files before installing.
What it does
Skillscript is a constrained declarative language, inspired by Make, and a runtime that turns an agent's reasoning into persistent automation. The agent writes a skill, a human approves what it may do, and it then runs the same way every time. It is built for teams that want agents to own recurring workflows without unrestricted shell access, package installs or direct control of production credentials. A skill is a typed workflow with variables, operations, dependencies and an output template. It can call connectors, branch, loop, run other skills and fire on a schedule or an HTTP event. The runtime refuses a `shell` call and a file write until the operator allowlists the binary and the path. In secured mode a new skill arrives as a draft and waits for approval, and approval signs the content with an operator key. The agent connects over MCP at `http://localhost:7878/rpc`, discovers the available tools, writes a skill and lints it. Runs start from the CLI, MCP, cron, an HTTP event or another skill. The runtime records traces, outputs and blocked operations.
Who it is for. Teams that want agents to own recurring processes and approve agent permissions without reading script code.
Good fit when
- An agent learned a procedure and it must be saved as an inspectable artifact instead of being re-derived each time
- A daily disk check or report runs on a schedule with extra commands denied
- Routine subtasks go to cheap local models while the expensive model writes the skill once
- Every skill passes static validation and manual approval before running
- Access to systems goes through connectors with no raw keys in the skill
Not a fit when
- You need a free-form Python or Bash program: the language is deliberately constrained and does not replace scripts
- You need a stable interface: the project is pre-1.0 and connector contracts are still settling
- The process is one-off: it is simpler for the agent to run it directly
Example request
Write a skill that takes `df -h` at six every morning and writes the result to a log, and send it to me for approval.Limitations
The project is pre-1.0: the core language and the connector contracts are stabilizing. The runtime installs globally through npm, and `skillfile init` creates the starting layout. Skill access to the shell and files is closed by default, and the operator opens it with allowlists of binaries and paths. The docs and the connector contract reference live on the project site. The dashboard starts with `skillfile dashboard`, and the MCP endpoint listens on port 7878 on localhost by default.
How to disable. Remove the package with `npm uninstall -g skillscript-runtime`, delete `http://localhost:7878/rpc` from the agent MCP config and remove the skills folder that `skillfile init` created.
FAQ
How do I connect Skillscript to an agent?
Run `skillfile dashboard` and add an MCP server at `http://localhost:7878/rpc`. The agent then sees the tools and can write skills.
How is a skill different from a Bash script?
A skill has no arbitrary imports, no `eval` and no subprocess escape, the shell and filesystem are closed by default, and side effects are visible on the page before a run.
Who approves a skill?
The human operator. Until consent is given, the skill stays an inert draft.
Where are credentials kept?
In the runtime, separate from the skill. Skills reach external systems through configured connectors.
What starts a skill?
A CLI command, an MCP call, a cron schedule, an HTTP event or another skill. Triggers such as `Triggers: cron:"0 6 * * *"` are set in the skill header.
Discussion
No questions or reviews yet.
Related
CLI toolsA skill that equips an agent with Kody CLI for local code execution using shared memory, packages and cloud account data.
Make Skills: skills for Make scenarios
Make Skills
Seven Make skills for AI agents: scenarios, running and debugging them, Functions v2, Endpoints and safe credential hand-off
Turns one run of a task into a program an agent calls as a single tool, checking the record before and after the save
A skill with references teaches an agent to install, configure and repair the OpenClaw gateway, channels and security.
Windows skills for agents on PowerShell: Outlook, Microsoft 365, Edge browser, desktop and system commands as structured JSON.
A skill set that teaches an agent to work with Gmail, Slack, HubSpot, Salesforce and other apps through the Membrane CLI
README in short
Covers the idea of the language, the comparison with Python and Bash, a sample skill, the author, review, run and observe loop, skill roles, connectors and the pre-1.0 status.