Getting it into your agent
There is no command for this one: it runs only inside a plugin, and the catalogue could not identify which plugin ships it. The source is linked below.
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.
[](https://agentmods.dev/instructions/raydestar/sir-thaddeus/harness-iteration-framework)<a href="https://agentmods.dev/instructions/raydestar/sir-thaddeus/harness-iteration-framework"><img src="https://agentmods.dev/badge/instructions/raydestar/sir-thaddeus/harness-iteration-framework.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 | $0.01634 | $0.01634 |
| Opus 5 | $0.00817 | $0.00817 |
| Sonnet 5 | $0.00327 | $0.00327 |
| Haiku 4.5 | $0.00163 | $0.00163 |
Grade A, and why
Harness Iteration Framework 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.
How it starts
The opening of the file, as written. The whole thing — 192 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Harness Iteration Framework
Use this workflow when the goal is to improve real product behavior through the Sir Thaddeus harness. Optimize for fast diagnosis, disciplined reruns, and general fixes.
Core Rule
Treat the harness as a measuring device, not the thing to optimize directly. Improve routing, tool use, synthesis, fallback behavior, state handling, or runtime transport in product code. Do not tailor behavior to known tests.
Phase 0: Pick The Fastest Useful Loop
Choose the narrowest loop that still gives trustworthy signal.
- Full baseline:
./dev/harness.ps1 --all --judge none - Suite baseline:
./dev/harness.ps1 --suite <name> --judge none - Single failing test:
./dev/harness.ps1 --suite <name> --test <id> --judge none - Judge validation after product fixes:
./dev/harness.ps1 --suite <name> --judge model
Rules:
- Start broad only when establishing current system state or confirming final regression coverage.
- After the first failing full run, narrow immediately to the failing suite or test.
- Do not keep rerunning the full harness while debugging one issue unless the failure suggests cross-suite state leakage or broad routing regressions.
Phase 1: Run The Full Baseline
When asked for the full harness state, run:
./dev/harness.ps1 --all --judge none
Capture:
- overall pass/fail summary
- failing suites/tests
- run id / artifact root
Do not conclude from the summary alone. The harness can under-report answer-quality defects when --judge none is used.
Phase 2: Break Failures Into Layers Before Editing
For each failing test, inspect artifacts in this order:
score.jsonfinal.txtsteps.jsonlinput.json
Decide which layer failed:
- No tools called when tools were needed: routing / intent classification / policy
- Wrong tools called: tool selection / route shaping
- Right tools called but wrong answer: synthesis / grounding / fallback logic
- Tool results missing or malformed: tool adapter / runtime transport / parsing
- Good answer but bad score: read the response and trace anyway; confirm whether this is a real product defect or a scoring blind spot
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 · 192 lines · 1,634 tokens per session scan A cb8a95d414f0
Harness Iteration Framework is an instructions file published in the GitHub repository raydeStar/sir-thaddeus (13 stars, last pushed 5d ago), licensed Apache-2.0. It adds 1,634 tokens to every session, about $0.0082 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
ironcurtain CLAUDE.md
Claude Code instructions for provos/ironcurtain, covering claude.md, general workflow, git & worktrees, git workflow and platform considerations.
TimeTree-MCP CLAUDE.md
Claude Code instructions for ehs208/TimeTree-MCP, covering timetree mcp project guide, project purpose, current capabilities, architecture map and api behavior notes.
llm-swarm-router AGENTS.md
AGENTS.md instructions for matthewdcage/llm-swarm-router, covering agent & developer guide, dox rail, project overview, architecture and repository layout.
llm-swarm-router CLAUDE.md
Claude Code instructions for matthewdcage/llm-swarm-router, covering claude code: swarm-llm and honcho memory (mcp).
mcp-local-llm AGENTS.md
Instructions for aplaceforallmystuff/mcp-local-llm, covering agents.md - mcp-local-llm, tech stack, architecture, development commands and environment variables.
abfall-mcp-server AGENTS.md
AGENTS.md instructions for AlpayC/abfall-mcp-server, covering notes for coding agents, setup, commands, the one principle and traps.