agentgrade
v1.2.0
Published
Install and manage the Agentgrade skill pack for Claude/Codex workflows.
Maintainers
Readme
agentgrade
agentgrade is an agent skill for Claude Code and Codex.
This npm package installs the skill into your local skills directory.
The grading workflow runs when you invoke the skill in chat with $agentgrade.
Requirements
- Node.js
>=20 - Codex or Claude Code
What the skill does
When you run $agentgrade, it is designed to produce:
- executive-style ranked actions (max 3) to improve developer workflow
- a clear decision prompt to choose plan, execute, or recommendations-only
- A-F gradecard
- hybrid confidence scoring (deterministic telemetry + semantic lane)
- instruction improvements
- MCP recommendations with shortlist-first lookup
Semantic judging is runtime-native: the active Codex/Claude agent is the intended semantic judge, with safe automatic fallback. No separate LLM setup is required.
Scope
agentgrade tracks workflow quality trends in your own agent sessions (retries, failures, token usage, safety signals) and suggests concrete improvements.
It does not attempt to be an external AI-news or model-release tracker.
Quick start
Install and activate the skill (recommended):
npx agentgrade installIf you already ran npm install agentgrade (or pnpm add agentgrade / yarn add agentgrade), still run:
npx agentgrade installnpm install only adds the package to node_modules. The install command is what copies the skill into your local Claude/Codex skills directories.
Use the skill:
- Restart Claude Code or Codex if it was already running.
- Open Claude Code or Codex in a repository.
- Run this in chat:
$agentgradeUninstall
npx agentgrade uninstallInstall location
The installer copies:
skills/agentgrade/SKILL.mdskills/agentgrade/bin/collect-evidence.mjsskills/agentgrade/bin/analyze-evidence.mjsskills/agentgrade/bin/semantic-judge.mjsskills/agentgrade/agents/openai.yamlskills/agentgrade/references/*
into:
~/.codex/skills/agentgrade~/.claude/skills/agentgrade
No environment variables are required for normal install/uninstall.
Additional docs
skills/README.md: standaloneskills/pack usage (manual copy/install distribution).skills/agentgrade/README.md: architecture and internals for theagentgradeskill.docs/ARCHITECTURE.md: top-down architecture, design reasoning, tradeoffs, and extension points.CHANGELOG.md: release notes by version.
