Eval Plan — Data skill for Claude Code
Decide what an LLM eval actually measures, before any code exists - gather the change and its call path from the repo, then settle six questions (goal, models, prompts, measure, guardrail, data) with.
How to install Eval Plan
Installs to ~/.claude/skills/oryxintel-oryxflow-claude-plugin-eval-plan/SKILL.md
mkdir -p ~/.claude/skills/oryxintel-oryxflow-claude-plugin-eval-plan && curl -fsSL https://raw.githubusercontent.com/oryxintel/oryxflow-claude-plugin/HEAD/commands/eval-plan.md -o ~/.claude/skills/oryxintel-oryxflow-claude-plugin-eval-plan/SKILL.md Restart Claude Code, or start a new session, for it to be picked up.
What Eval Plan does
description: Decide what an LLM eval actually measures, before any code exists - gather the change and its call path from the repo, then settle six questions (goal, models, prompts, measure, guardrail, data) with a proposal drawn from what is really there. Writes only evals//README.md, the durable record of what is measured and why. Run it when you changed a prompt and cannot tell whether it is better. disable-model-invocation: true
Plan an eval: decide what is measured, write n
Alternatives in Data
- LLM Redteam — LLM red-team corpus runner — fires categorized prompt-injection / jailbreak / system-prompt-leak / data-exfil 4.5k ★
- Sxgen — Generate Protheus SX data dictionary configuration scripts - SX3 fields, SIX indexes, SX1 questions, SX5 gener 136 ★
- TikHub API Helper — TikHub API Helper is a Claude Code Agent Skill that helps users search, discover, and call TikHub APIs 62 ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Lane Ingest
Data ingest lane. Own every reader of a surface the game writes - the log parser and tail, the GVAS save reade
OSS Migration Eval
Decide whether to switch an LLM pipeline to a cheaper/open-source model — honest cost-vs-accuracy verdict with
Sec Request
Call any SEC Open Data API path, including GET/POST v1 endpoints and cursor-paginated v2 endpoints
Data Check
Check data exists and meets quality standards before training
Redis Agentcore Settle Agent
Debt settlement agent for a fictional bank: Claude on Amazon Bedrock AgentCore Runtime with Redis Cloud as the
AI Agents Irreversible Guard
Two-tier guardrail for AI coding agents: hard-stop irreversible actions (rm -rf, DROP DATABASE, deleting a clo
Related Agents
LLM Evaluator
LLM evaluation specialist. Use when designing eval suites for LLM features, calibrating LLM-as-judge prompts,
BI Doc Generator
Generates and updates model and report documentation (measure dictionary, table overview, report page guide) u
AI Prompt Engineer
Desk Studio's AI Prompt Engineer. Writes and fixes prompts for image, video and audio generators the user has: