sir-thaddeus: Instructions file for GitHub Copilot

.github/instructions/local-model-harness-workflow.instructions.md

Local-Model Harness Tuning Workflow is an instructions file for GitHub Copilot from raydeStar/sir-thaddeus. It costs 4,932 tokens per session, scanned C, original, Apache-2.0.

An operating guide for testing a local language model with an evaluation harness. A language model is software that generates text or code; the harness measures how well it performs tasks.

In plain words
What is it for?
Verifying local model settings, confirming the model server is using the expected model, establishing a baseline, investigating failures, and iterating on harness results.
Why use it?
It reduces misleading results caused by using the wrong model, skipping a baseline, or making unfocused changes. It describes a repeatable loop for checking, diagnosing, changing, and retesting.

Instructions file for GitHub Copilot

Written for GitHub Copilot: a Copilot instructions file.

This is raydeStar/sir-thaddeus's own configuration. It tells GitHub Copilot how to work on sir-thaddeus itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything sir-thaddeus configures →

Reuse

Borrowing it

Nothing to install: this file belongs to raydeStar/sir-thaddeus. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/raydeStar/sir-thaddeus/master/.github/instructions/local-model-harness-workflow.instructions.md
Clone the repo
git clone --depth 1 https://github.com/raydeStar/sir-thaddeus

Made for: GitHub Copilot.

Wrote 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.

agentmods badge for Local-Model Harness Tuning Workflow

README.md
[![agentmods](https://agentmods.dev/badge/instructions/raydestar/sir-thaddeus/local-model-harness-workflow.svg)](https://agentmods.dev/instructions/raydestar/sir-thaddeus/local-model-harness-workflow)
Your own site
<a href="https://agentmods.dev/instructions/raydestar/sir-thaddeus/local-model-harness-workflow"><img src="https://agentmods.dev/badge/instructions/raydestar/sir-thaddeus/local-model-harness-workflow.svg" alt="Measured on agentmods" height="20"></a>
Per session 4,932 This file is loaded in full into every session.
When invoked 4,932 The same file — it is already loaded in full.
Security scan C 2 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.04932 $0.04932
Opus 5 $0.02466 $0.02466
Sonnet 5 $0.00986 $0.00986
Haiku 4.5 $0.00493 $0.00493

Measured 7d ago against content hash b500a14cfa94, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade C, and why

Local-Model Harness Tuning Workflow scanned grade C with 2 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 7d 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.

Downloads and executes remote codehighSupply chain

curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.

curl -s http://localhost:1234/v1/models | python -m json.tool

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

curl -s http://localhost:1234/v1/models | python -m json.tool
.github/instructions/local-model-harness-workflow.instructions.md · 311 lines

How it starts

The opening of the file, as written. The whole thing — 311 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Local-Model Harness Tuning Workflow

Companion to e2e-harness-rules.instructions.md. Those rules stay in force — this file documents the operational loop: verify model, baseline, triage, patch, iterate. Written from a live session where we went from 25% pass rate to stable 54%+ with targeted fixes.

Read this BEFORE running the harness on a new model size. Most of the friction this session came from skipping a step below.


0. Sanity-check the model before anything else

Changing the model in LM Studio (or llama.cpp, or Ollama, etc.) is not enough. The runtime reads llm.model from:

%LOCALAPPDATA%\SirThaddeus\settings.json

Verify from PowerShell:

(Get-Content "$env:LOCALAPPDATA\SirThaddeus\settings.json" -Raw | ConvertFrom-Json).llm |
  Select-Object model, gatekeeperModelId, contextWindowTokens, maxTokens

Then confirm the LM Studio server actually has that id loaded:

curl -s http://localhost:1234/v1/models | python -m json.tool

If the response's data[].id doesn't match llm.model, either flip the settings or the LM Studio alias. Wrong model = wrong baseline = every fix you write is chasing a ghost.

1. Baseline before you patch

Always run the full baseline on the target model once before you change a single line of code. Failures on a 4B look very different from failures on a 2B — budgets, routing, and keyword coverage all shift.

# Stop any stale runtime/MCP first (the harness forks its own; stale ones eat GPU)
Stop-Process -Name 'Thaddeus.Runtime','SirThaddeus.McpServer' -Force -ErrorAction SilentlyContinue

# Single suite, quick sanity
.\dev\harness.ps1 --suite smoke --judge none

# Full baseline (30-90 min on 4B, 10-20 min on 2B)
.\dev\harness.ps1 --all --judge none

Capture pass/fail numbers per suite before you touch anything:

$suites = @('smoke','reasoning','web-search','quality','tool-contracts','personality','footman-validation','existence')
foreach ($s in $suites) {
  $out = .\dev\harness.ps1 --suite $s --judge none 2>&1 | Out-String
  $p = [regex]::Match($out, 'Passed: (\d+)').Groups[1].Value
  $f = [regex]::Match($out, 'Failed: (\d+)').Groups[1].Value
  "$s`t$p/$(([int]$p + [int]$f))"
}

Read the full file on GitHub · 311 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. 7d ago First seen · 311 lines · 4,932 tokens per session scan C b500a14cfa94

Subscribe to this mod's changes

Local-Model Harness Tuning Workflow is an instructions file published in the GitHub repository raydeStar/sir-thaddeus (13 stars, last pushed 6d ago), licensed Apache-2.0. It adds 4,932 tokens to every session, about $0.0247 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other instructions, from other repositories

next.js AGENTS.md

AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.

vercel/next.js · 7,296 tokens

codex AGENTS.md

AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.

openai/codex · 5,182 tokens

vscode buildNext.instructions.md

Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).

microsoft/vscode · 6,785 tokens

vscode oss-third-party-notices.instructions.md

Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).

microsoft/vscode · 5,001 tokens

langchain AGENTS.md

AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.

langchain-ai/langchain · 4,469 tokens

spec-kit AGENTS.md

AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.

github/spec-kit · 7,104 tokens