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.
npx agentmods add instructions/agentevalhq/agenteval/assertionsgit clone --depth 1 https://github.com/AgentEvalHQ/AgentEvalWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00629 | $0.00629 |
| Opus 5 | $0.00315 | $0.00315 |
| Sonnet 5 | $0.00126 | $0.00126 |
| Haiku 4.5 | $0.00063 | $0.00063 |
Grade A, and why
AgentEval assertions.instructions.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.
How it starts
The opening of the file, as written. The whole thing — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Assertion Implementation Guidelines
Core Assertion Pattern
All assertions follow the fluent pattern with chainable methods:
result.ToolUsage!.Should()
.HaveCalledTool("Tool1")
.BeforeTool("Tool2")
.WithArgument("key", "value")
.And() // Return to parent assertions
.HaveNoErrors();
Adding New Assertion Method
- Add to appropriate assertions class:
ToolUsageAssertions,PerformanceAssertions, orResponseAssertions - Always add
[StackTraceHidden]attribute to keep stack traces clean - Accept optional
becauseparameter for documentation - Use
AgentEvalScope.FailWith()for rich error messages
Assertion Method Template
/// <summary>Assert that XYZ condition is met.</summary>
/// <param name="expected">The expected value.</param>
/// <param name="because">Optional reason for the assertion.</param>
[StackTraceHidden]
public ToolUsageAssertions HaveXyz(string expected, string? because = null)
{
if (!ConditionMet(expected))
{
AgentEvalScope.FailWith(
ToolAssertionException.Create(
$"Expected XYZ to be '{expected}', but it was not.",
expected: expected,
actual: _actualValue,
suggestions: new[] { "Try doing X", "Check Y" },
because: because));
}
return this;
}
Error Message Structure
All failure messages MUST include:
- Expected: What was expected
- Actual: What was found
- Suggestions: Actionable hints (when applicable)
- Because: User's reason (if provided)
Example output:
Expected tool 'SearchTool' to be called because query requires search.
Expected: Tool 'SearchTool' called at least once
Actual: Tools called: [CalculateTool, FormatTool]
Suggestions:
→ Verify the agent has access to the expected tools
→ Check if the prompt clearly requests tool usage
Exception Types
ToolAssertionException- Tool-related assertion failuresPerformanceAssertionException- Performance/cost assertion failuresResponseAssertionException- Response content assertion failuresAgentEvalScopeException- Multiple failures collected in scope
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.
- yesterday First seen · 88 lines · 629 tokens per session scan A ca5cdbc35aed
AgentEval assertions.instructions.md is an instructions file published in the GitHub repository AgentEvalHQ/AgentEval (138 stars, last pushed yesterday), licensed MIT. It adds 629 tokens to every session, about $0.0031 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.
Other instructions, from other repositories
zeroclaw CLAUDE.md
Instructions for zeroclaw-labs/zeroclaw, covering claude.md — zeroclaw (claude code), claude code settings, hooks and slash commands.
openagent CLAUDE.md
Claude Code instructions for the-open-agent/openagent, covering claude.md, commands, architecture, backend (go / beego) and frontend (react).
Tracely-ai CLAUDE.md
Claude Code instructions for Jwuthri/Tracely-ai, covering claude.md, commands, architecture, hard rules and gotchas.
nuwax AGENTS.md
Instructions for nuwax-ai/nuwax, covering ai agent system documentation, 系统概述, ai agent 架构, 核心组件 and ai 功能特性.
zhin zhin-plugin.instructions.md
Instructions for zhinjs/zhin, covering zhin plugin runtime authoring, package contract, convention directories, imports and native typescript and command routes.
OpenPersona AGENTS.md
Instructions for acnlabs/OpenPersona, covering agents.md, project overview, setup, project structure and architecture rules.