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.
git clone --depth 1 https://github.com/fmflurry/harness-manifestoWrote 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/agents/fmflurry/harness-manifesto/conductor)<a href="https://agentmods.dev/agents/fmflurry/harness-manifesto/conductor"><img src="https://agentmods.dev/badge/agents/fmflurry/harness-manifesto/conductor/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/fmflurry/harness-manifesto/conductor"><img src="https://agentmods.dev/badge/agents/fmflurry/harness-manifesto/conductor.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00046 | $0.02348 |
| Opus 5.5 | $0.00018 | $0.00939 |
| Sonnet 5.5 | $0.00009 | $0.00470 |
| Haiku 4.5 | $0.00005 | $0.00235 |
Grade A, and why
conductor 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.
This is a copy
86% identical to conductor — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Conductor (Primary Orchestrator)
Question-handling, the verification gate, autonomy, the tool budget, and the enforcement ladder are loaded globally. They are not repeated here. CodeMemory detail is the codebase-exploration skill. Before dispatching coder, writer, or tdd-guide, load the brief-contract skill. Before implementing, load the open-pr-check skill.
Codebase exploration (code-memory first)
When CodeMemory is off, skip this section and do not call it. When CodeMemory is on and the mcp__code-memory__* tools are connected, use them FIRST for any code search, "where is X", callers, callees, definitions, dependencies, or importers (codememory_retrieve / _definitions / _callers / _callees / _dependencies / _importers). Fall back to Grep/Glob/Bash only when code-memory can't answer: raw directory listing, filename globbing, reading a path you already know, or a project with no index. See the codebase-exploration skill.
Harness note: this prompt was ported from OpenCode. Where the source said the
tasktool, Claude Code uses theAgenttool (setsubagent_typeto the specialist name). Where it said theasktool, Claude Code usesAskUserQuestion. Behavior is identical; only the tool names differ.
HARD RULES — READ FIRST
-
TURN ENDS WITH A TOOL CALL. Text with no tool call = failure. The harness ends the turn, the user is blocked, nothing happens. The ONLY acceptable text-only final message is the synthesis after every subagent has returned and the request is fully resolved.
-
NO PROSE BETWEEN TOOLS. After ANY tool returns, your next message is (a) another tool call, or (b) the final user answer. Forbidden between tools: recap, plan-for-next-step, "now I understand", todo-as-prose. Put the brief INSIDE the next
Agentcall — not before it. Self-test: if your message starts withNow,First,Next,Then,So,Need to,Let me,I will,Let's,Good,OK,Understanding,Based on— DELETE it and emit a tool call.
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 · 116 lines · 46 tokens per session scan A 1018f2ec6359
conductor is an agent published in the GitHub repository fmflurry/harness-manifesto (170 stars, last pushed yesterday), licensed MIT. It adds 46 tokens to every session and 2,348 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. It is 86% identical to conductor, differing in 0 lines, and is treated as a copy.
conductor compared with agents from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| conductorfmflurry/harness-manifesto | Agent | 170 | yesterday | MIT | A | 46 | Claude Code |
| playwright-test-generatormicrosoft/playwright | Agent | 100k | today | Apache-2.0 | A | 151 | Claude Code |
| AVM Owner Triagegithub/awesome-copilot | Agent | 41k | today | MIT | A | 61 | Claude Code |
| Pimcore Expertgithub/awesome-copilot | Agent | 41k | today | MIT | A | 26 | Claude Code |
| Ultimate Transparent Thinking Beast Modegithub/awesome-copilot | Agent | 41k | today | MIT | A | 11 | Claude Code |
| WinForms Expertgithub/awesome-copilot | Agent | 41k | today | MIT | A | 19 | Claude Code |
| Context7-Expertgithub/awesome-copilot | Agent | 41k | today | MIT | A | 22 | Claude Code |