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/sbroenne/mcp-server-excel/llm-testing-philosophygit clone --depth 1 https://github.com/sbroenne/mcp-server-excelWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/sbroenne/mcp-server-excel/llm-testing-philosophy)<a href="https://agentmods.dev/instructions/sbroenne/mcp-server-excel/llm-testing-philosophy"><img src="https://agentmods.dev/badge/instructions/sbroenne/mcp-server-excel/llm-testing-philosophy.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.03451 | $0.03451 |
| Opus 5 | $0.01725 | $0.01725 |
| Sonnet 5 | $0.00690 | $0.00690 |
| Haiku 4.5 | $0.00345 | $0.00345 |
Grade A, and why
mcp-server-excel llm-testing-philosophy.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 6d 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.
Copies of this mod
1 near-identical copy found in the catalogue:
- mcp-server-ppt llm-testing-philosophy.instructions.md — 89% identical, 138 lines differ
How it starts
The opening of the file, as written. The whole thing — 381 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LLM Testing Philosophy
⚠️ CORE PRINCIPLE: Tests simulate real users. Failures expose product gaps, not test gaps.
What Are LLM Tests?
LLM tests use an AI coding agent (with its own default system prompt) to exercise our CLI and MCP tools. The agent receives a natural language prompt — the same kind a real user would type — and must figure out how to accomplish the task using our tools.
These tests do NOT test the LLM. They test our product's usability surface:
- Skill documentation (SKILL.md)
- CLI
--helpoutput - MCP tool descriptions (XML
/// <summary>) - Error messages and recovery hints
- Parameter naming and discoverability
- Workflow coherence
The Golden Rule
If the LLM can't figure it out, fix the product — never fix the test.
When an LLM test fails, the root cause is ALWAYS one of:
- Our skill docs don't explain the workflow clearly enough
- Our tool descriptions are misleading or incomplete
- Our CLI --help output doesn't show the right examples
- Our error messages don't guide recovery
- Our parameter names are confusing or undiscoverable
- The test itself is unreasonable (rare — only fix if a human couldn't do it either)
What NEVER Belongs in a Test
❌ xfail or skip Markers
NEVER use @pytest.mark.xfail or @pytest.mark.skip to hide failing tests.
Tests either pass or fail. There is no middle ground.
xfailmasks real failures and creates a false sense of progressskiphides broken code instead of fixing it- If a test fails, fix the product — the test is exposing a real problem (Golden Rule)
- If a test is flaky, fix the flakiness — don't paper over it with xfail
❌ CLI Command Guidance in Prompts
A real user doesn't know our CLI syntax. Neither should the test prompt.
# ❌ WRONG: Teaching the LLM how to use our CLI
prompt = """
Create a PivotTable then set layout to Compact using
'excelcli pivottablecalc' (run --help to see options).
The compact layout uses row-layout value 0.
"""
# ✅ CORRECT: Natural user request
prompt = """
Create a PivotTable with Compact layout showing
Department and Team as rows, Hours as values.
"""
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.
- 6d ago First seen · 381 lines · 3,451 tokens per session scan A 59b92a5a2acd
mcp-server-excel llm-testing-philosophy.instructions.md is an instructions file published in the GitHub repository sbroenne/mcp-server-excel (656 stars, last pushed yesterday), licensed MIT. It adds 3,451 tokens to every session, about $0.0173 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
mcp-server-powerpoint copilot-instructions.md
Copilot instructions for sbroenne/mcp-server-powerpoint, covering project instructions — powerpointmcp, critical files (read these first), sister projects, what is powerpointmcp? and session model (mcp server).
mcp-server-powerpoint mcp-server-guide.instructions.md
Instructions for sbroenne/mcp-server-powerpoint, covering mcp server development guide, llm-facing content rules, two kinds of tools: hand-written vs. generated, implementation pattern: single hand-written dispatch tool and error handling (mandatory).
mcp-server-powerpoint architecture-patterns.instructions.md
Instructions for sbroenne/mcp-server-powerpoint, covering architecture patterns, .net class design (mandatory), layered architecture (critical), command pattern and structure (per domain).
mcp-server-powerpoint critical-rules.instructions.md
Instructions for sbroenne/mcp-server-powerpoint, covering critical rules - must follow, rule 1: success/errormessage invariant, rule 1b: no exception suppression in core, rule 30: real-com integration tests only, strict tdd and rule: 1-based indexing everywhere.
mcp-server-powerpoint testing-strategy.instructions.md
Instructions for sbroenne/mcp-server-powerpoint, covering testing strategy, strict tdd (red → green), test project layout, serialization is mandatory and trait tagging (required for surgical filtering).
mcp-server-amazon CLAUDE.md
Instructions for rigwild/mcp-server-amazon, covering claude.md, development commands, install dependencies (use -d flag for puppeteer), build typescript to javascript and clean mock html files.