Skillscript
A small, strict recipe language for AI agents — instead of letting an agent freely improvise a routine task every single time (and possibly do something risky), you write the steps once as a checked recipe it has to follow.
🔗 Visit Skillscript
Description
AI agents are good at figuring out a task from scratch, but that's also the problem: ask the same agent to do the same routine job twice — say, checking overnight support tickets — and it might reason its way there differently each time, burning tokens and occasionally doing something unexpected. Skillscript is like handing the agent a laminated checklist instead of asking it to improvise from memory each time: once a task is written as a "skill," it runs the same deterministic way every time, with guardrails on what it's allowed to touch.
Technically, Skillscript is a declarative language (syntax inspired by Make) plus a runtime, compiler, and CLI for authoring agent workflows that execute through configured "connectors" to external systems, with security controls including shell/filesystem allowlists, credential management, and a human-approval step before a new skill can run unsupervised. It integrates with MCP (Model Context Protocol), supports cron scheduling and event triggers, includes static validation before execution, and logs audit trails for everything a skill does. It's MIT licensed, open source, free, and still pre-1.0.
💬 Our review
The short version: this is aimed at a real and growing problem — agents re-deriving the same routine task expensively and unpredictably — but as a pre-1.0, single-maintainer project it's a bet on an idea more than a mature tool yet.
Against LangChain or established orchestration frameworks, Skillscript's angle is narrower and more disciplined: it's not trying to be a general agent-building framework, it's specifically a way to freeze a proven agent routine into something deterministic and auditable, with security allowlists baked in rather than bolted on. Against n8n or Zapier, which orchestrate integrations via visual workflows for humans, Skillscript is explicitly built for agents authoring their own skills, with a human-approval gate rather than a human building the whole flow. The catch is maturity: pre-1.0, MIT, and thin on independent validation means you should expect the language and APIs to shift, and the security controls — the actual selling point — haven't had much outside scrutiny yet.
📊 Global score
🤖 AI-enriched data
MIT, entièrement gratuit, pas d'offre payante.
Pros
Gratuit, open source, ciblé sur un vrai problème
Contrôles de sécurité intégrés dès la conception
Exécution déterministe des tâches répétitives
Intégration MCP, cron, audit trails
Cons
Pré-1.0, API/langage susceptibles de changer
Projet à échelle d'un seul mainteneur
Contrôles de sécurité peu éprouvés en externe
Nouvelle syntaxe à apprendre