Skip to content

AGENTS.md

Latest commit

 

History

History
77 lines (53 loc) · 1.83 KB

AGENTS.md

File metadata and controls

77 lines (53 loc) · 1.83 KB
Raw

AGENTS.md

Project

agenthint is a TypeScript CLI/library for detecting when a tool is probably being run by an AI agent. Detection is advisory only and must not be treated as a security boundary.

Shell

Prefix shell commands with rtk unless the command genuinely needs raw shell behavior.

Examples:

rtk npm run check
rtk git status
rtk sed -n '1,200p' src/index.ts

Commands

Use mise from the repo root for toolchain versions:

mise install
mise exec -- npm run check

Use these project scripts:

npm run build
npm run format
npm run format:check
npm run lint
npm run test
npm run check
cargo clippy --workspace --all-targets -- -D warnings
cargo test --workspace

Run npm run check before committing or handing off code changes.

If editing install.sh, run sh -n install.sh and keep it POSIX sh compatible.

Code Style

  • TypeScript source lives in src/.
  • Rust source lives in crates/agenthint/src/.
  • Tests live in test/ and use Node's built-in test runner.
  • Keep detection results explainable: include confidence and signals.
  • Prefer explicit AI_AGENT support over heuristics.
  • Never print environment variable values that may contain secrets; signal names are enough.
  • Keep filesystem probes documented and configurable.

CLI Behavior

The CLI should preserve these exit codes:

  • 0: agent runtime likely detected
  • 1: agent runtime not detected
  • 2: invalid usage or detection error

Setup-only commands such as agenthint init <agent> should exit 0.

agenthint doctor should explain what was detected and recommend setting AI_AGENT when detection is heuristic.

Git

Use Conventional Commits, for example:

feat: add parent process detection
fix: avoid printing sensitive env values
docs: expand agent integration guide

Husky runs npm run check before commits.