Skill Grader — Development agent for Claude Code
Grades a run's outputs against its assertions, quoting the output for every grade, so a grade without evidence can't pass.
How to install Skill Grader
Installs to ~/.claude/agents/thenavidm-skill-builder-skill-grader.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/thenavidm/skill-builder/HEAD/agents/skill-grader.md -o ~/.claude/agents/thenavidm-skill-builder-skill-grader.md Restart Claude Code, or start a new session, for it to be picked up.
What Skill Grader does
name: skill-grader description: Grades a run's outputs against its assertions, quoting the output for every grade, so a grade without evidence can't pass. Started by skill-eval and skill-compete.
You grade one run. You get the run's folder (its `answer.md` and the files it wrote) and the task's assertions.
For each assertion:
- If code can check it (a file exists, contains or matches something, is valid JSON, stays under a length), check it exactly and say what you found.
- If it need
Alternatives in Development
- Engram Assessor — Independent grader of learner productions for the Engram learning plugin 1.4k ★
- Dashclaw Gate Runner — Runs the DashClaw verification gates (lint, full vitest suite, build, contract checks) and returns ONLY the fa 297 ★
- Shep Port Creator — Creates ONE brand-new output port interface in packages/core/src/application/ports/output/ without any caller 247 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Skill Analyzer
Finds the patterns in eval and bench results that averages hide, assertions that never tell the skill apart, t
Command Grader
Grades command execution transcripts against assertion definitions (standard, hard-gate, contract)
Grader Hiring Manager
One seat of the resume panel. Grades a resume against a job description from the hiring manager's lens (can th
Grader Recruiter
One seat of the resume panel. Grades a resume against a job description from the recruiter's lens (the 6-secon
Learn Assessor
Blind grader for the learning system (teach / tutor / level-up). Grades ONE learner answer against a frozen, v
Us Performance Grader
US stock performance grader. Grades past picks on the v6.3 RANK bar (BIG WIN = weekly rank ≤ 100, WIN = ≤ 500,
Related Skills
Performance Marketing Eval
15 performance marketing tasks for evaluating AI agents, with simulated tools, a shared grader, and baseline t
Claim Grade
Run the AI claim grader on a piece of text — commit message, PR body, or sprint section
Grokkable Output
Agent skill that makes AI replies parseable in one pass — verdict first, prose over fake structure, claims tha