Entire

You are reviewing an IMPLEMENTATION PLAN (not code yet) for the Entire CLI repo (Go, cwd is the repo root). Be a skeptical senior reviewer. Read the actual source files the plan names and verify the plan is correct, feasible, and idiomatic for THIS codebase. Focus on: correctness of the technical approach (PID + start-time process-liveness detection), cross-platform pitfalls (the darwin sysctl kinfo_proc decoding and linux /proc/<pid>/stat field parsing — note field 2 comm can contain spaces/parens), import-cycle risk, whether it reuses existing helpers, race/safety concerns with mutating session state, and any missing edge cases. Point out anything wrong or risky and suggest concrete fixes. Be concise — bullet points, most important issues first. Do not write or edit files.

Here is the plan:

I don’t see the implementation plan after the colon, so I’m going to inspect the relevant liveness/session-state code paths first and then call out the missing-plan blocker if there isn’t enough to review concretely.

The repo already has a split between cmd/entire/cli/session for persisted state primitives and cmd/entire/cli/strategy wrappers, plus a lockfile package. I’m reading those now because any PID/start-time design has to fit those boundaries without creating cycles or racy read-modify-write behavior.

State saves are atomic renames but not compare-and-swap, so any plan that “load, mutate, save”s session JSON from concurrent hook processes needs a lock or a very narrow write path. I’m checking the hook lifecycle entry points to see where PID fields would be set and cleared.

One important thing in the code shape: normalized lifecycle events don’t currently carry an agent process PID. If the plan intends to infer it from the hook process itself, that would only prove the short-lived entire hooks ... subprocess was alive, not the agent session.