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/herdrdev/herdr/agents-mdgit clone --depth 1 https://github.com/herdrdev/herdrWhat 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 | $0.04985 | $0.04985 |
| Opus 5 | $0.02492 | $0.02492 |
| Sonnet 5 | $0.00997 | $0.00997 |
| Haiku 4.5 | $0.00498 | $0.00498 |
Grade A, and why
herdr 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 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.
How it starts
The opening of the file, as written. The whole thing — 291 lines — stays where its author put it; the contents beside it link to each section on GitHub.
herdr
Terminal based agent runtime for coding agents.
Scope and Audience
These instructions are layered.
- Unless a section explicitly says it is maintainer-only, local-machine-only, or external-contributor-only, treat it as universal project guidance.
- Universal project rules apply to every agent working on Herdr, including forks.
- Maintainer accounts are listed in
.github/MAINTAINERS. Treat the acting account as a verified maintainer only when its username is listed there, the configured remote is the canonicalherdrdev/herdrrepository, and the authenticated account has write access to that repository. If any condition cannot be verified, skip maintainer workflow and follow the external contributor guardrail instead. - Local Can machine workflow applies only on Can's own workstation or Windows
VM setup, for example when
/home/can/Projects/herdr,HERDR_ENV=1, or thewindows-wirtSSH alias exists. If those facts are not true, skip local machine workflow. - External contributor guardrail applies whenever the acting GitHub account is not a verified maintainer, the work is happening in a fork, or the account cannot be determined.
Universal Project Rules
Principles
- State is separated from runtime.
AppStateis pure data, testable without PTYs or async.PaneStateis separate fromPaneRuntime. Workspace logic doesn't need real terminals. - Render is pure.
compute_view()handles geometry and mutations.render()takes&AppStateand only draws. Never mutate state during render. - No god objects. If a module is doing too many things, split it.
app/is already split into state, actions, and input. Keep it that way. - Platform code is isolated. OS-specific behavior lives in the matching
src/platform/<os>.rsfile, with only shared traits, types, wrappers, and testable contracts insrc/platform/mod.rs. Core modules don't have#[cfg(target_os)]. - Detection is decoupled. The detector reads a screen snapshot, never touches the parser or viewport state.
- Screen detection is evidence-based. When changing
src/detect/manifests/, first capture the relevant bottom-buffer state withherdr agent read <pane> --source detection --format textand, when styling or alternate screen behavior matters,--format ansi. Decide which visible controls are invariant, which are alternatives, and encode them as explicit AND/OR gates. Do not match whole-pane incidental text, and do not use the user-visible viewport for agent status because users can scroll it. - UI patterns should be reused. Herdr is a mouse-first TUI. New dialogs, onboarding, settings, and post-update flows should follow the existing UI/UX language and interaction patterns instead of inventing one-off screens. Prefer reusing existing modal/screen structure, affordances, and close actions so the app feels consistent.
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 · 291 lines · 4,985 tokens per session scan A 27d2768d62b9
herdr AGENTS.md is an instructions file published in the GitHub repository herdrdev/herdr (33,954 stars, last pushed today), licensed Apache-2.0. It adds 4,985 tokens to every session, about $0.0249 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
kungfu AGENTS.md
Instructions for kungfu-systems/kungfu, covering agents.md, start from the person's objective, rules that always apply, load bounded context before changing the repository and execute and verify through shifu.
jarvis-registry AGENTS.md
Instructions for ascending-llc/jarvis-registry, covering jarvis registry — agent guide, project overview, ⚠️ plan mode instructions ⚠️, my engineering preferences and 1. architecture review.
kungfu CLAUDE.md
Instructions for kungfu-systems/kungfu: Read and follow AGENTS.md before working in this repository. It routes both product use and contribution work to the repository sources of truth, including the requirement that development and build tasks enter through Shifu.
kungfu copilot-instructions.md
Instructions for kungfu-systems/kungfu: Read and follow AGENTS.md before proposing or running changes. It is the shared agent router for this repository, including the canonical Shifu development and build entrypoint.
luvus AGENTS.md
Instructions for RizRiyz/luvus, covering agents.md, start with current evidence, product model, non-negotiable development safety and architecture and code map.
beevibe CLAUDE.md
Instructions for beevibe-ai/beevibe, covering beevibe — claude code guide, monorepo structure, running locally, critical: @beevibe/core compiles to dist/ and restarting cleanly.