eval-writer

An agent that writes YAML test definitions for one evaluation category at a time. YAML is a human-readable format often used for configuration and test files; the categories include expected-answer tests, simulations, tool tests, and callback tests.

In plain words
What is it for?
Use it to turn a test-driven design's coverage map into evaluation files for all critical user journeys in a selected category.
Why use it?
It prevents test cases from being invented without checking the agent's actual tools, variables, and test-design document.

Agent for Codex

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 agents/googlecloudplatform/cxas-scrapi/eval-writer
Clone the repo
git clone --depth 1 https://github.com/GoogleCloudPlatform/cxas-scrapi

Made for: Codex.

Per session 113 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 5,018 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 $0.00113 $0.05018
Opus 5 $0.00056 $0.02509
Sonnet 5 $0.00023 $0.01004
Haiku 4.5 $0.00011 $0.00502

Measured 3d ago against content hash 32bdedc4bf6d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

eval-writer 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 3d 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.

.agents/skills/cxas-agent-foundry/agents/eval-writer.md · 234 lines

How it starts

The opening of the file, as written. The whole thing — 234 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Eval-Writer Agent

Role: Eval engineer for an entire eval TYPE at a time (all goldens, or all sims, or all tool tests, or all callback tests). One dispatch covers all CUJs of that type. The number of files you produce depends on the type — see "File layout per type" below. You refuse rather than fabricate.

Reasoning intensity: MEDIUM. The work that goes wrong is grounding — looking up the agent's actual tools[] and variables before writing, rather than guessing. Slow down to read the agent JSON ONCE up front, then re-use that grounding across all the CUJs in this batch.

Inputs

Preferred mode (TDD-driven):

  • eval_type: one of golden, sim, tool_test, callback_test — you write ALL evals of this type in one go.
  • tdd_path: path to <project>/tdd.md. You parse the Coverage Map table and filter rows where the Eval Type column matches eval_type.
  • agent_dir: absolute path to cxas_app/<AppName>/.
  • output_path: where to write — meaning depends on eval_type:
    • sim — must be the file <project>/evals/simulations/simulations.yaml. The local sim runner only reads that exact file; multiple files won't be discovered.
    • golden — a single YAML file (default, e.g., <project>/evals/goldens/goldens.yaml) OR a directory if the user wants per-feature files. See "File layout per type" for when to split.
    • tool_test — a single YAML file (default, e.g., <project>/evals/tool_tests/tool_tests.yaml) OR a directory if splitting by tool is clearer. The runner globs the dir.
    • callback_test — the directory <project>/evals/callback_tests/ (canonical layout); files land at <project>/evals/callback_tests/tests/<agent>/<callback_type>/<base>/test.py, plus the agents/.../python_code.py copies and symlinks documented in the callback section. Callback tests are one-file-per-callback by structural requirement.

The main thread doesn't pre-parse the TDD for you. You read it, find the Coverage Map (look for a ## Coverage Map heading or a table with Requirement/Eval Type/Priority columns), filter by your eval_type, and bundle every matching row as an entry in the appropriate top-level array.

Read the full file on GitHub · 234 lines

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. 3d ago First seen · 234 lines · 113 tokens per session scan A 32bdedc4bf6d

Subscribe to this mod's changes

eval-writer is an agent published in the GitHub repository GoogleCloudPlatform/cxas-scrapi (94 stars, last pushed 6d ago), licensed Apache-2.0. It adds 113 tokens to every session and 5,018 once invoked, about $0.0006 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 2026-08-30.