AGENTS.md

The AGENTS.md capability reads project instruction files from the session workspace and injects them into the agent’s system prompt at the start of every turn. By default it reads AGENTS.md, Everruns’ implementation of the AGENTS.md open standard — an emerging convention backed by OpenAI, Google, Cursor, Sourcegraph, and the Linux Foundation.

When the capability is enabled:

• The agent reads /workspace/AGENTS.md on every turn by default.
• Agents can configure files to read additional workspace files such as CLAUDE.md.
• Edits during a session apply on the next turn (no restart).
• The content is prepended to the system prompt, before capability fragments and the agent’s own prompt.
• If a configured file doesn’t exist, the agent operates normally.

Prompt order

When AGENTS.md is present, the system prompt is composed top-to-bottom:

1. Instruction file content — project-level instructions
2. Capability system prompt additions — tool guidance
3. Agent’s base system prompt — role and personality

Project context comes first, so downstream prompt fragments can build on it.

Limits

• Content cap: 32 KiB per file. Excess is silently truncated.
• Plain Markdown — no required sections.

Compatibility

Everruns keeps AGENTS.md as the default. Configure the capability with files to read additional tool-specific files:

{
  "files": ["AGENTS.md", "CLAUDE.md"]
}

Do something

• Use AGENTS.md for project instructions — full guide with examples.
• Equip an agent with tools — adding capabilities generally.

See also

• AGENTS.md capability reference — config schema.