This repository contains standalone external agent binaries that extend the Entire CLI with support for additional AI coding agents.

External agents are standalone binaries (named entire-agent-<name>) that teach Entire CLI how to work with AI coding agents it doesn't natively support. When an external agent is installed on your PATH, Entire discovers it automatically and gains the ability to:

• Create checkpoints during AI coding sessions so you can rewind mistakes
• Capture transcripts of what the AI agent did and why
• Install hooks so the AI agent's lifecycle events (start, stop, commit) flow through Entire

External agents communicate with Entire CLI via subcommands that accept and return JSON over stdin/stdout. See the external agent protocol spec for the full interface.

See each agent's own README for setup and usage instructions.

This repo includes a skill that guides you through building a new external agent with two test layers:

1. Protocol compliance — generic subcommand coverage from entireio/external-agents-tests
2. Lifecycle integration — repo-local e2e/ tests that exercise entire enable, prompt execution, checkpoints, and rewind
3. Implementation — build the binary until both layers pass, then add unit tests

Clone the repo and open it in your AI coding tool. Each tool auto-discovers the skill with no additional configuration:

The skill files live in .claude/skills/entire-external-agent/ if you want to read the details.

Testing is intentionally split:

• Generic protocol checks run in GitHub Actions via entireio/external-agents-tests. The workflow builds each entire-agent-* binary in this repo and runs the shared compliance suite against it.
• Lifecycle tests stay in this repo's e2e/ harness. These verify the parts that depend on Entire itself and on the real agent CLI: prompt execution, hook installation after entire enable, checkpoint creation, rewind behavior, and interactive sessions.
• Unit tests live with each agent implementation under agents/.

# Run unit tests for all agents
mise run test-unit

# Run lifecycle integration tests from this repo
mise run test-e2e

# Same as test-e2e, kept as the explicit name
mise run test-e2e-lifecycle

# Run unit + lifecycle tests locally
mise run test-all

Protocol compliance runs in CI through .github/workflows/protocol-compliance.yml.

The lifecycle harness auto-discovers and builds all agents in agents/ via TestMain:

agents/                          # Standalone external agent projects
  entire-agent-kiro/             # Kiro agent (Go binary)
  entire-agent-pi/               # Pi agent (Go binary)
e2e/                             # Lifecycle integration harness
.github/workflows/               # CI, including protocol compliance via external-agents-tests
.claude/skills/entire-external-agent/  # Skill files (research, test-writer, implementer)
AGENTS.md                        # Codex auto-discovery
.cursor/rules/                   # Cursor auto-discovery
.opencode/plugins/               # OpenCode auto-discovery