design-agent-eval-workflow

A planning framework for testing whether an agent, skill, prompt, or automation behaves correctly. An evaluation workflow defines test cases, pass criteria, and when a human must review the result.

In plain words
What is it for?
Use it to choose evaluation cases, graders, thresholds, regression runs, automation levels, and escalation points before implementation.
Why use it?
It turns vague confidence in agent behavior into repeatable checks and identifies cases that should stop for review, such as failed validation or unsafe side effects.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/gaelic-ghost/socket/design-agent-eval-workflow
Any agent
npx skills add gaelic-ghost/socket --skill design-agent-eval-workflow
Clone the repo
git clone --depth 1 https://github.com/gaelic-ghost/socket

Made for: Claude Code, Codex.

Per session 68 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,267 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00068 $0.01267
Opus 5 $0.00034 $0.00633
Sonnet 5 $0.00014 $0.00253
Haiku 4.5 $0.00007 $0.00127

Measured 2d ago against content hash 1f61f5add3fd, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

design-agent-eval-workflow scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/agent-engineering-skills/skills/design-agent-eval-workflow/SKILL.md · 121 lines

How it starts

The opening of the file, as written. The whole thing — 121 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Design Agent Eval Workflow

Design evaluation workflows for agent, skill, prompt, and automation behavior.

This skill is a framework-neutral eval planning surface. It helps define what good behavior means, how to measure it repeatably, which tasks can safely run fully automatically, and where escalation remains necessary because safety cannot be engineered into the workflow with reasonable confidence.

Inputs

  • Required: the agent, skill, prompt, automation, or workflow to evaluate
  • Useful: target runtime, task examples, expected outputs, failure modes, write surface, external services touched, acceptable risk, and current validation commands
  • Optional: preferred eval runner, existing case format, model/provider constraints, budget, cadence, and reporting destination

Workflow

  1. Restate the behavior under evaluation and the real decision the eval should support.
  2. Define the automation target:
    • full-auto: safe to run without human approval when it passes objective gates
    • auto-with-escalation: runs automatically but stops on bounded ambiguity, failed checks, secret exposure, destructive writes, or external side effects
    • human-review: requires review because the task cannot yet be made safe enough through scope, validation, sandboxing, rollback, or orchestration
    • manual-only-for-now: behavior is too underspecified or high-risk for a useful eval-backed automation claim
  3. Build the case set:
    • happy path
    • realistic messy input
    • regression examples from past failures
    • refusal or stop conditions
    • tool, filesystem, network, or credential boundary cases
    • full-auto eligibility cases when automation is the goal
  4. Choose graders:
    • deterministic assertions for structure, file changes, commands, outputs, and policy boundaries
    • snapshot review only when stable text shape matters
    • model grading only for judgment that cannot be reduced to deterministic checks, with explicit rubrics and sampled audit review
  5. Set pass thresholds, stop conditions, and escalation rules.
  6. Choose the run surface:
    • local script or uv run pytest for repo-local skill and prompt evals
    • codex exec or Codex GitHub Action for repeatable repo tasks
    • OpenAI Agents SDK tracing/evals when application code owns tools, handoffs, guardrails, and traces
    • LangGraph evaluation or LangSmith-style tracing when a stateful graph owns transitions, persistence, resume behavior, or graph-level observability
    • a stack-owned test runner when the workflow belongs to a narrower plugin
  7. Produce a scaffold with cases, grader shape, thresholds, reporting, automation eligibility, and implementation handoff.
  8. Link official docs for every framework, runtime, or eval surface named.

Read the full file on GitHub · 121 lines

Files

What ships with it

3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 121 lines · 68 tokens per session scan A 1f61f5add3fd

Subscribe to this mod's changes

design-agent-eval-workflow is a skill published in the GitHub repository gaelic-ghost/socket (7 stars, last pushed 7d ago), licensed Apache-2.0. It adds 68 tokens to every session and 1,267 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

use-gemini-web

Use the user's logged-in Gemini Web session as an untrusted auxiliary model when its public YouTube audio/visual understanding, public-URL context, or an independent second analysis materially helps Codex answer or verify a request. Use for one or multiple public videos, follow-up questions, cross-checking, or a fresh…

andrewLi1994/gemini-web-bridge · 110 tokens

create-skill

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

warpdotdev/warp · 64 tokens

factory-files

Create and edit file-based Warp software factory definitions, in a repository tree rooted at a factory.yaml. Use when authoring or changing that factory.yaml, Agent, Automation, Scorer, Runner, or Webhook files under that root, or its factory and agent skill trees, and when fixing Factory file diagnostics. Do not use…

warpdotdev/warp · 101 tokens

skill-management

Create, edit, and delete custom managed skills in the user's workspace. Use whenever the user wants to author a new skill from a description, scaffold a SKILL.md, or remove a skill they no longer need.

vellum-ai/vellum-assistant · 46 tokens

archon

Autonomous multi-session campaign agent. Decomposes large work into phases, delegates to sub-agents, reviews output, and maintains campaign state across context windows. Use for work that spans multiple sessions and needs persistent state, quality judgment, and strategic decomposition.

SethGammon/Citadel · 54 tokens

create-app

End-to-end app creation from a single description. Five tiers: blank project, guided, templated, fully generated, or feature addition to existing codebase. Routes through PRD, architecture, and Archon campaign with verification at every step.

SethGammon/Citadel · 52 tokens