Add agents.txt v1.0 at repo root

[barneywohl]

Add agents.txt (v1.0) at repo root

Hey — first off, thanks for the work on this repo. We shipped @agent_press/mcp-server yesterday on top of the MCP TypeScript SDK, and a lot of what makes that possible is the clarity these reference implementations provide. This PR is meant in that spirit: a small, peer contribution from a project that builds on MCP.

What this PR adds

A single file: agents.txt at the repository root.

agents.txt is a new open standard (v1.0, shipped this week) that lets a repository declare, in a machine-readable INI format, what autonomous AI agents are permitted to do when interacting with the project. It sits in the same lineage as robots.txt, sitemap.xml, and llms.txt — advisory, static, no runtime component. Spec: https://github.com/barneywohl/agentpress/blob/main/docs/AGENTSTXT_SPEC.md

The version proposed here is intentionally conservative for a monorepo of reference implementations:

• allowed_actions covers reading, running tests in CI, opening PRs, and commenting on issues
• prohibited_actions covers merging, publishing packages, modifying SECURITY.md / LICENSE / CODE_OF_CONDUCT.md, adding new reference servers, and any protocol-spec changes
• requires_human_approval flags anything under src/**, .github/workflows/**, dependency changes, and behavior changes to documented tool schemas
• entry_points points to existing README.md, CONTRIBUTING.md, and CLAUDE.md
• scope mirrors CONTRIBUTING.md's existing preference for small, focused PRs and explicitly marks new servers and protocol changes as out of scope

What this PR does not do

• No CI changes, no workflow changes, no new dependencies
• No changes to CONTRIBUTING.md, SECURITY.md, or any existing file
• No protocol-spec implications — agents.txt is purely a repo-level declaration
• Does not attempt to enumerate per-server rules; the file is monorepo-root scope only

Caveats I want to flag

agents.txt at the root of a monorepo has real subtleties — different subdirectories have different test commands, different maintainers, and different risk profiles. The version here errs on the side of being restrictive (broad requires_human_approval patterns, conservative scope limits) rather than precise. Per-server agents.txt files could be added later if useful, but that's out of scope for this PR.

Happy to iterate on wording, tighten the prohibited_actions list, or close this if it's not a fit for the repo's direction.