AGENTS.md Context [agent_md]

Complete config.toml to disable automatic repository context:

[agent_md]
enabled = false

An omitted table inherits; an authored table replaces the lower table and uses defaults for omitted fields. Unknown fields fail. See config.toml.

Repository-authored context is untrusted input. Review it before running in an unfamiliar workspace; admission is independent of tool read permissions. See security boundaries.

Controls root-run AGENTS.md discovery documented in Agents. This is the global default; an agent document may set agent_md: true or agent_md: false to override it for that agent, for example to keep a lean agent's context free of repository instructions:

---
description: Minimal-context agent
mode: primary
enabled: true
models:
  - { model: "openai/gpt-5", variant: null }
agent_md: false
---
Answer using only what the user provides.

Key	Type	Default	Description
enabled	boolean	true	Load AGENTS.md context files at each root run start. Delegated and internal agents remain excluded. An agent document's agent_md field, when present, overrides this for that agent's root runs.

Files larger than 2 MiB (2097152 bytes) are skipped entirely and surfaced as an AgentMdSkipped warning event; they are never partially loaded or truncated. Loaded entries are rendered into a single user context turn wrapped in a <system-reminder> block, with each file delimited as <contents from="/abs/path/AGENTS.md">…</contents>.