CtrlK
BlogDocsLog inGet started
Tessl Logo

mcollina/init

Creates, updates, or optimizes an AGENTS.md file for a repository with minimal, high-signal instructions covering non-discoverable coding conventions, tooling quirks, workflow preferences, and project-specific rules that agents cannot infer from reading the codebase. Use when setting up agent instructions or Claude configuration for a new repository, when an existing AGENTS.md is too long, generic, or stale, when agents repeatedly make avoidable mistakes, or when repository workflows have changed and the agent configuration needs pruning. Applies a discoverability filter—omitting anything Claude can learn from README, code, config, or directory structure—and a quality gate to verify each line remains accurate and operationally significant.

85

1.14x
Quality

94%

Does it follow best practices?

Impact

72%

1.14x

Average score across 5 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

criteria.jsonevals/scenario-5/

{
  "context": "An agent is given an existing AGENTS.md alongside source files (.cursorrules and .github/copilot-instructions.md) and asked to update the AGENTS.md. The task mentions that agents keep using the wrong test command (npm test instead of npm run test:unit). The ideal response reads the source files, applies a discoverability filter (only adding non-obvious guidance agents can't get from reading public files like package.json), fixes root causes where possible rather than accumulating workarounds, removes stale content, and produces a shorter, more focused AGENTS.md.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Consults .cursorrules",
      "description": "The agent reads or references .cursorrules before producing the updated AGENTS.md, rather than working only from the existing AGENTS.md and the task description.",
      "max_score": 8
    },
    {
      "name": "Consults .github/copilot-instructions.md",
      "description": "The agent reads or references .github/copilot-instructions.md before producing the updated AGENTS.md.",
      "max_score": 8
    },
    {
      "name": "Incorporates non-discoverable guidance from source files",
      "description": "The updated AGENTS.md includes at least one piece of non-obvious guidance drawn from .cursorrules or copilot-instructions.md that was not already in the original AGENTS.md (e.g., server-component-first rule, app/api/ routing convention, db client import rule, experimental/ directory warning).",
      "max_score": 10
    },
    {
      "name": "Does not blindly copy all cursor rules content",
      "description": "The agent applies a discoverability filter and does NOT wholesale copy every bullet from .cursorrules into AGENTS.md. Generic, easily discoverable items (e.g., 'This is a Next.js 14 app') are omitted or paraphrased at most; only agent-relevant, non-obvious rules are included.",
      "max_score": 6
    },
    {
      "name": "Recommends root-cause fix for wrong test command",
      "description": "Rather than (or in addition to) adding a note to AGENTS.md, the agent recommends or applies a root-cause fix — such as renaming/aliasing the npm test script in package.json so `npm test` runs test:unit, or adding a lint/CI rule. A note-only response with no mention of tooling fixes scores zero.",
      "max_score": 12
    },
    {
      "name": "Removes stale Docker setup instruction",
      "description": "The Docker / docker-compose setup instruction (which the existing AGENTS.md itself flags as possibly obsolete since early 2025) is removed from the updated AGENTS.md rather than retained or merely re-annotated.",
      "max_score": 10
    },
    {
      "name": "Retains the production migration caveat",
      "description": "The manual `npm run db:migrate:prod` step after production deployment is present in the updated AGENTS.md. This non-obvious, high-stakes step appears in both the existing AGENTS.md and copilot-instructions.md and must be preserved.",
      "max_score": 10
    },
    {
      "name": "Does not include generic discoverable guidance",
      "description": "The updated AGENTS.md omits guidance that any developer could trivially discover from README, package.json, or tsconfig (e.g., 'run npm install', 'we use ESLint and Prettier', 'TypeScript strict mode') — these do not belong in AGENTS.md under the discoverability principle.",
      "max_score": 6
    },
    {
      "name": "Output is shorter or more focused than existing AGENTS.md",
      "description": "The updated AGENTS.md has fewer total lines or words than the original, or is demonstrably more concise, reflecting removal of stale/redundant content and application of the discoverability filter.",
      "max_score": 6
    },
    {
      "name": "Shows incremental update rather than blind replacement",
      "description": "The agent preserves the structural intent of the existing AGENTS.md (e.g., retains section headings that remain relevant, keeps wording that is still accurate) rather than discarding it entirely and rewriting from scratch. The output reads as an evolution of the original.",
      "max_score": 6
    },
    {
      "name": "Correct test command present in updated AGENTS.md",
      "description": "The updated AGENTS.md explicitly specifies `npm run test:unit` for unit tests (not `npm test`) regardless of whether a root-cause tooling fix is also recommended.",
      "max_score": 8
    },
    {
      "name": "Next.js App Router / server component guidance included",
      "description": "At least one of the key architectural rules from .cursorrules is present in the updated AGENTS.md — specifically either the server-component-first rule or the app/api/ routing convention (not the legacy pages/api/ pattern). This is agent-relevant, non-obvious, and not discoverable from standard docs.",
      "max_score": 10
    }
  ]
}

evals

SKILL.md

tile.json