boshu2__agentops/docs/MIGRATION-3.0.md

Migrating to AgentOps 3.0

3.0 is a deliberate narrowing. AgentOps is now the in-session agent operating loop and context compiler: skills, the rpi/evolve loops, crank/swarm, and the .agents/ corpus. Everything that tried to be an always-on orchestration runtime has been removed. Out-of-session orchestration is delegated to a substrate you choose. AgentOps adopts a reference trio — NTM (a local tmux swarm) + MCP (ao mcp serve, shipped) + managed-agents (ao agent) — or you can run Olympus (the full-custom take). AgentOps owns none of it.

This guide lists what was removed in 3.0 and what to use instead.

Removed: all hooks (the repo is hookless)

ao hooks install, --with-hooks, and the embedded hook surface are gone. AgentOps no longer installs or runs hooks.

Use instead: skills + the ao CLI guide the workflow; CI is the authoritative gate (.github/workflows/validate.yml). What a PreToolUse/PostToolUse hook used to enforce locally, a CI job now enforces on push. For the in-session knowledge loop, the skills (/rpi, /evolve, /post-mortem) call the ao commands directly instead of relying on hook side effects.

Removed: the daemon (ao daemon / agentopsd)

The always-on daemon — the queue, supervisor, scheduler, job executors, and the dream/overnight compounding runner — is deleted. AgentOps has no out-of-session runtime.

Use instead:

• In-session: run the loop yourself — ao rpi (one cycle), ao evolve (many cycles), crank/swarm for in-session agent teams. This is the zero-dependency path; it needs no extra binary.
• Out-of-session (always-on): run the loop under a substrate AgentOps adopts but does not own — an NTM tmux swarm (workers each run ao rpi <bead>; a lead agent or operator slings ready beads), MCP via ao mcp serve (exposes the ao tool surface to any MCP-aware harness), or managed-agents via ao agent (hosted, scheduled drivers). The substrate supplies supervision, crash-recovery, and agent spawning; the agents inherit the AgentOps skills and run the AgentOps loops.
• Or use Olympus for a full-custom out-of-session implementation.

Removed: scheduling (ao schedule, ao plans, ao watch, ao overnight)

These drove the daemon's cron lane and are gone with it. docs/scheduling.md and the .agents/schedule.yaml contract no longer apply.

Use instead: schedule the substrate you adopt — cron-triggered NTM dispatch or a managed-agents schedule that runs ao compile / ao maturity / corpus maintenance — or run the work in-session on demand.

Removed: the factory command (ao factory) and its contract corpus

Use instead: the factory is the loop — crank/swarm in-session, or substrate-driven dispatch (NTM / managed-agents) out-of-session. There is no separate factory binary.

Removed: the runtime=gc phased-engine bridge

The CLI gc-bridge glue is severed. ao rpi keeps its non-gc backends (auto/direct/stream/tmux); runtime=gc is no longer a valid mode. A substrate now dispatches whole loops (ao rpi) as one unit rather than driving the phased engine through a bridge.

New in 3.0

• ao validate --gate — deterministic exit-code verdict (0 pass/warn, 1 fail, 2 internal error; --strict flips WARN→1). The retry hook for a substrate's check step and for CI; composes the existing ratchet validator, no network or LLM.
• The reference out-of-session substrate — AgentOps adopts NTM + MCP (ao mcp serve, shipped) + managed-agents (ao agent) to run the loop out of session. None of it is AgentOps-owned; the substrate dispatches a whole ao rpi/ao evolve loop as one unit and never drives its insides.

What stays (the in-session core)

ao rpi, ao evolve, crank, swarm, ao harvest, ao forge, ao mine, ao compile, ao wiki, ao doctor, ao goals, and the .agents/ corpus are all unchanged. The in-session loop is the product, and it is proven end-to-end.

One-line summary

If you ran AgentOps in a session, nothing you rely on changed. If you ran the daemon for always-on work, move that lane to a substrate (NTM + MCP + managed-agents) or Olympus — AgentOps itself is now in-session only.