Problem: Long OMP sessions produce cognitive loops — “optimizing the wrong problem”, attachment to results (Gita), fear vs clarity, or drifting from long-term goals. Full retros (clock-aligned or break-time) cost too much context; need lightweight 30-second reorientation nudges injected as steer messages.

Solution: @ankitg12/agent-suggester Pi extension. Loads via omp plugin link (not marketplace). Registers ExtensionAPI factory that subscribes to session_start, tool_call, before_agent_start, agent_end, session_shutdown. Schedules setTimeout(fire, intervalMs) on every session start (session-relative by default).

Install (ghq + link):

ghq get https://github.com/ankitg12/agent-suggester-omp
omp plugin link $(ghq list ankitg12/agent-suggester-omp)

Manifest in package.json:

"pi": { "extensions": ["./agent-suggester.ts"] }

Config (~/.omp/agent/agent-suggester.json):

{
  "intervalMs": 900000,
  "clockAlign": false,
  "picker": "random",
  "skipIfIdle": true,
  "wrapWith": "🔄 **Suggestion** — a 30-second pause invited\n\n{item}\n\n*If this shifts something, say so and we'll adjust. Otherwise continue — no analysis needed.*",
  "items": [
    "Are you solving the right problem, or optimizing the wrong one?",
    ...
    "This too shall pass."
  ]
}

Scheduling & injection:

Evidence (2026-06-21 session):