evals AGENTS.md

A practical guide for AI agents working on a benchmark of React Native coding tasks. It explains how the runner discovers tasks, generates solutions, and uses a language model to judge the results.

In plain words
What is it for?
Running evaluations, generating solution artifacts, judging completed tasks, inspecting per-task results, and reviewing summary reports.
Why use it?
It clarifies the repository’s execution process and required reasoning steps, helping agents produce results that can be evaluated consistently.

Instructions file for CodexOpenCode

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 instructions/callstackincubator/evals/agents-md
Clone the repo
git clone --depth 1 https://github.com/callstackincubator/evals

Made for: Codex, OpenCode.

Per session 1,309 This file is loaded in full into every session.
When invoked 1,309 The same file — it is already loaded in full.
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 $0.01309 $0.01309
Opus 5 $0.00655 $0.00655
Sonnet 5 $0.00262 $0.00262
Haiku 4.5 $0.00131 $0.00131

Measured yesterday against content hash 6f7611c3d54d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evals AGENTS.md 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 yesterday.

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.md · 137 lines

How it starts

The opening of the file, as written. The whole thing — 137 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS Guide

This file is the practical playbook for agents working in this repo.

What this repo is

  • Purpose: benchmark how coding models solve React Native tasks.
  • Primary engine: runner/ orchestrates discovery -> solve -> LLM judge -> summary.
  • Dataset: evals under evals/<category>/<eval-id>/, typically with:
    • prompt.md
    • requirements.yaml
    • app/ (baseline input)
    • reference/ (reference output for judge context)

Mental model of execution

Benchmark execution uses two CLIs:

  1. bun runner/run.ts discovers evals and generates artifacts under the configured output directory.
  2. bun runner/judge.ts reads generated artifacts, runs LLM judging, and writes results under results/<run-id>/.

Generation details (runner/solver/pipeline.ts):

  • --model is required and is always used for generation.

Judge details (runner/evaluators/llm/run.ts):

  • --model is required and LLM judge always runs against generated artifacts.

Key output behavior:

  • Per-eval results: results/<run-id>/evals/<eval-id>.json
  • Summary: results/<run-id>/summary.json
  • --debug adds judge prompt/output artifacts.

Think-before-coding rules (required)

Apply these before implementation:

  1. State assumptions explicitly.
  2. If multiple interpretations exist, list them and ask or choose with justification.
  3. If a simpler approach works, use it and say why.
  4. If something is unclear, stop and surface the exact ambiguity.

Execution principles (default policy)

Before implementing

  • State assumptions explicitly. If uncertain, ask.
  • If multiple interpretations exist, present them; do not pick silently.
  • If a simpler approach exists, call it out and use it unless there is a clear reason not to.

Simplicity and scope control

  • Implement only what the task asks for.
  • No speculative abstractions for single-use code.
  • No drive-by refactors.
  • Match surrounding style.
  • Remove only dead code created by your own change.
  • If you notice unrelated issues, mention them but do not fix unless asked.
  • Remove imports/variables your changes made unused; do not clean pre-existing dead code unless requested.

Read the full file on GitHub · 137 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. yesterday First seen · 137 lines · 1,309 tokens per session scan A 6f7611c3d54d

Subscribe to this mod's changes

evals AGENTS.md is an instructions file published in the GitHub repository callstackincubator/evals (110 stars, last pushed 1mo ago), licensed MIT. It adds 1,309 tokens to every session, about $0.0065 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.