eval-view AGENTS.md

Project instructions for EvalView, a tool that tests AI agents, records their results, and compares later runs with saved baselines. A baseline is a recorded expected result used to spot regressions, or unwanted changes.

In plain words
What is it for?
Use them when developing EvalView, adding or changing test and evaluation behavior, saving snapshots, checking regressions, or reviewing terminal and HTML differences.
Why use it?
They explain the project’s test cases, evaluation results, review process, and rules for detecting changes in agent output, tool use, cost, speed, and safety checks.

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/hidai25/eval-view/agents-md
Clone the repo
git clone --depth 1 https://github.com/hidai25/eval-view

Made for: Codex, OpenCode.

Per session 2,634 This file is loaded in full into every session.
When invoked 2,634 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 $0.02634 $0.02634
Opus 5 $0.01317 $0.01317
Sonnet 5 $0.00527 $0.00527
Haiku 4.5 $0.00263 $0.00263

Measured 2d ago against content hash 279ee3f6e0fe, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

eval-view 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 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.

AGENTS.md · 422 lines

How it starts

The opening of the file, as written. The whole thing — 422 lines — stays where its author put it; the contents beside it link to each section on GitHub.

EvalView Agent Instructions

This file is written for coding agents working inside the EvalView repository. Use it as the fastest way to understand the architecture, the important invariants, and the common extension paths.

For the repo's product and development philosophy, also read:

  • docs/OPERATING_MODEL.md

What EvalView Is

EvalView is a regression gate for AI agents.

Its core job is:

  1. Load test cases.
  2. Execute them against an agent backend.
  3. Evaluate outputs, tool use, cost, latency, and safety-related checks.
  4. Save baselines with snapshot.
  5. Compare current behavior against baselines with check.
  6. Render reviewable terminal and HTML diffs.

EvalView is not just a scorer. It is opinionated around baseline comparison, regression detection, and developer-facing review flows.

Internally, the intended loop is:

  1. specify the behavior change
  2. implement with agent help
  3. review hard
  4. run the relevant behavior-tagged slice
  5. snapshot intentional deltas
  6. run check
  7. ship

Core Concepts

TestCase

Path: evalview/core/types.py

TestCase is the declarative input spec. It includes:

  • the input prompt or turns
  • expected tools and sequences
  • thresholds
  • optional contains / not_contains / regex / schema checks
  • optional per-test adapter and endpoint overrides

If you are changing test-loading or authoring behavior, inspect:

  • evalview/core/types.py
  • evalview/core/loader.py

EvaluationResult

Path: evalview/core/types.py

EvaluationResult is the evaluated output of one executed test. It contains:

  • passed
  • score
  • evaluations
  • trace
  • timestamps and metadata

This is the main object passed into reports, baselines, and diffing helpers.

GoldenStore

Path: evalview/core/golden.py

GoldenStore persists and loads baselines and variants from .evalview/golden/.

Use it for:

  • saving baselines during snapshot
  • loading all variants during check
  • counting and managing variants

DiffEngine

Read the full file on GitHub · 422 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. 2d ago First seen · 422 lines · 2,634 tokens per session scan A 279ee3f6e0fe

Subscribe to this mod's changes

eval-view AGENTS.md is an instructions file published in the GitHub repository hidai25/eval-view (133 stars, last pushed 9d ago), licensed Apache-2.0. It adds 2,634 tokens to every session, about $0.0132 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.