Herdforge AGENTS.md

An agent-work contract for the Herdforge software factory, including engineering rules, package ownership, and a required fleet-based workflow.

In plain words
What is it for?
Use it when coordinating Herdforge tasks, dispatching fleet agents, reviewing changes, checking package ownership, and running the project's verification workflow.
Why use it?
It tells participating agents who may perform each kind of work and how failures, reviews, commits, and approvals must be handled.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/kampe/herdforge/agents-md
Clone the repo
git clone --depth 1 https://github.com/Kampe/Herdforge

Made for: Codex, OpenCode.

Per session 2,630 This file is loaded in full into every session.
When invoked 2,630 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
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 $0.02630 $0.02630
Opus 5 $0.01315 $0.01315
Sonnet 5 $0.00526 $0.00526
Haiku 4.5 $0.00263 $0.00263

Measured 2d ago against content hash 3f6a4996b2b4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

Herdforge AGENTS.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 2d 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.

AGENTS.md · 140 lines

How it starts

The opening of the file, as written. The whole thing — 140 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Herdforge Agentic Worker Contract

This contract governs all AI agents (Claude, Codex, Grok, AGY, and OpenCode) participating in the Herdforge software factory fleet.


1. Core Engineering Invariants & Rules

  1. Fleet-Only Forging (FAC-113, hard invariant):

    • Herdforge forges itself with its own Herdr fleet (configured vendor harnesses in Herdr tabs, driven by herd dispatch/herd send/herd review). This is the entire point of the repo: the product must be exercised on itself.
    • NEVER use Claude subagents (the Agent tool: general-purpose, Explore, etc.) to build, fix, review, or audit anything in this repo. Using them defeats the dogfooding premise and stops testing the actual system.
    • The coordinator dispatches to the fleet and owns git plumbing (harvest → PR → merge → herd approve). Fleet agents commit in their worktree and report; they never push or merge. When an agent stalls, the fix is a firmer packet / herd kick / the completion self-gate (herd verify), never routing around the fleet.
  2. Fail-Closed Execution:

    • All CLI tools and internal package functions must propagate non-zero exit codes on error (os.Exit(1) or non-nil error).
    • HTTP responses carrying {"error": ...} JSON bodies returned under 200 OK must be rejected as hard errors.
  3. Mutation & Test Non-Vacuity Discipline:

    • Never write a test that passes vacuously.
    • Any negative assertion guard must be manually or programmatically verified failing against an introduced regression before PR approval.
  4. Repo-Relative Path Enforcement:

    • Never output absolute file paths (e.g. /Users/... or /tmp/...) inside configuration files, git hooks, or generated artifacts.
    • Use ./ or runtime environment variables ($HERD_ROOT, $HERD_WORKTREE) to ensure worktree portability.
  5. Deterministic Task Selection:

    • When listing candidate tasks from Kaneo, GitHub, Linear, Jira, or Azure DevOps, candidate arrays must be sorted deterministically: $$\text{Sort Order} = (\text{Priority DESC}, \text{Ticket Number ASC})$$
    • Agents must check that role labels match their assigned persona (pulse --role <role>) before claiming work.

Read the full file on GitHub · 140 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. 2d ago First seen · 140 lines · 2,630 tokens per session scan A 3f6a4996b2b4

Subscribe to this mod's changes

Herdforge AGENTS.md is an instructions file published in the GitHub repository Kampe/Herdforge (5 stars, last pushed 2d ago), licensed MIT. It adds 2,630 tokens to every session, about $0.0132 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 2026-08-31.