agents AGENTS.md

A set of instructions for coding agents working on agents, a self-hosted Go service that dispatches AI command-line tools to work on GitHub repositories. It describes the dashboard, APIs, event triggers, isolated runner containers, and agent-to-agent dispatching.

In plain words
What is it for?
Use it when changing the Go daemon, dashboard, REST or MCP interfaces, GitHub event handling, scheduled jobs, runner containers, or inter-agent dispatch logic.
Why use it?
It gives agents the project's operating model and boundaries, including how repositories are selected and how work runs in fresh containers. This context helps them follow the repository's architecture and workflow.

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/eloylp/agents/agents-md
Clone the repo
git clone --depth 1 https://github.com/eloylp/agents

Made for: Codex, OpenCode.

Per session 5,779 This file is loaded in full into every session.
When invoked 5,779 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 $0.05779 $0.05779
Opus 5 $0.02890 $0.02890
Sonnet 5 $0.01156 $0.01156
Haiku 4.5 $0.00578 $0.00578

Measured 2d ago against content hash 6112343146a6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agents 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 · 178 lines

How it starts

The opening of the file, as written. The whole thing — 178 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md

Repo-specific guidance for any coding agent (Claude Code, Codex, Cursor, Aider, …) working in this project. Cross-references CLAUDE.md for tool-specific notes and docs/local-models.md for the local-model integration story.

What this repo is

agents is a self-hosted Go daemon that dispatches AI CLIs (Claude Code, Codex, or any CLI pointed at a local LLM via the built-in proxy) to work on GitHub repos. The fleet lives in SQLite and is managed through the dashboard, REST, MCP, or YAML import/export. Agents are bound to repos via labels, GitHub event subscriptions, or cron schedules, and execute inside fresh ephemeral runner containers, which in turn use GitHub MCP tools first and gh fallback for complex local checkout/test/PR loops. The daemon itself is strictly read-only against GitHub.

Agents can also invoke each other at runtime via the reactive inter-agent dispatcher: an agent returns dispatch: [{agent, number, reason}] in its response JSON, the daemon validates against per-agent whitelists and safety limits, then enqueues a synthetic agent.dispatch event that runs the target agent.

Key numbers:

  • Language: Go 1.25 (check go.mod).
  • Binary entrypoint: cmd/agents/main.go.
  • Single-binary daemon deployment; the daemon image is the control plane, while the separate runner image includes the AI CLIs plus git, gh, Go, Rust/Cargo, Node/npm, TypeScript, and Flutter/Dart. GitHub access flows through MCP first, with gh as fallback.

Quick commands

go test ./...          # normal development / PR test loop
go test ./... -race    # full race pass, normally left to main-branch CI
docker compose pull    # pull the published image
docker compose up -d   # run the daemon

The supported runtime is Docker Compose, there is no local-binary workflow. On-demand runs go through the running daemon: POST /run (HTTP) or the trigger_agent MCP tool. There is no separate CLI mode for ad-hoc execution, it would be a second runtime that doesn't share the daemon's run-lock or dispatch dedup, opening a memory-write race window.

Read the full file on GitHub · 178 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 · 178 lines · 5,779 tokens per session scan A 6112343146a6

Subscribe to this mod's changes

agents AGENTS.md is an instructions file published in the GitHub repository eloylp/agents (9 stars, last pushed 4d ago), licensed MIT. It adds 5,779 tokens to every session, about $0.0289 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.