Tessl is an AI developer platform and agent orchestration harness built to develop, evaluate, and optimize agent skills and coding workflows. It provides a structured CLI environment for managing repository skills, generating test scenarios, and systematically benchmarking agent performance against standardized developer tasks. In daily use, Tessl operates through the `tessl` CLI and integrates directly with agent runtimes such as Claude Code. Users can install skill packages, run static reviews of instruction files (`SKILL.md`), automatically generate multi-scenario evaluation suites, and execute differential evaluations with and without the skill active using models like Claude 3.7 Sonnet, Opus, or Haiku. Unlike traditional CLI agents that focus solely on executing single-turn code generation tasks, Tessl functions as an agent evaluation harness. It tests whether an agent correctly routes to a skill via activation evals, scores output deltas using LLM-as-a-judge rubrics, and categorizes findings into a four-bucket framework to pinpoint performance gaps, redundant context, and instruction-induced regressions.
Tags: ai agent, cli, code assistant, coding agent, developer tools, Generative AI
SKILL.md files against completeness, actionability, conciseness, workflow clarity, and trigger term conflict risks. - Automated Scenario Generation — Generates realistic, scoped real-world task scenarios from skill specifications with tessl scenario generate. - Differential Baseline Evals — Runs tasks with and without skill context to measure precise score deltas and isolate net performance improvements. - Skill Activation Routing Testing — Identifies whether agents route to the intended skill using --solver=activation and suggests description rewrites for routing failures. - Four-Bucket Diagnostic Framework — Categorizes eval criteria into Working Well, Plugin Gap, Redundant, and Regression to streamline prompt and instruction optimization. - Multi-Model Agent Execution — Evaluates skill performance across specific target LLM backends including Claude Sonnet, Opus, and Haiku.Software engineers, platform teams, and AI tool builders developing and distributing skills, plugins, and custom instructions for AI coding agents. It is not intended for developers looking for a basic interactive terminal chat tool without evaluation and skill management needs.
bash tessl i tessl-labs/skill-optimizerbash tessl loginbash # Run static analysis on skill instructions tessl skill review skills/fastify/SKILL.md # Test if the agent routes to the correct skill tessl eval run . --solver=activation # Generate 5 test scenarios and benchmark with Claude Sonnet tessl scenario generate . --count=5 tessl eval run . --agent=claude:claude-sonnet-4-6Part of CLI Coding Agents
Categories: AI Coding Agents, Code Assistants, Developer Tools