ah-ydx Stage 3: mem0 memory lifecycle + Context Pack epic Lined Up

Per SPEC SS14.3: mem0 REST client (X-API-Key, scoping user_id=proj:<slug>/global, run_id=task:<id> — agent_id filter is buggy upstream); load cascade on claim -> .task/CONTEXT.md section; save on Done; post-run summarizer role; Context Pack builder (task+thread+parent-chain+memory+git, ~40k char budget, deterministic order, truncate-from-tail).

Audit note (2026-07-20): the SPEC 14.3 'post-run summarizer role' was never implemented as a role — the shipped design substitutes the agent-authored .task/summary.md + mem0 infer=true LLM extraction at Done-time (memory.go). All four ah-ydx children are closed.

DECISION STILL NEEDED — analysis added 2026-08-05 to make it a one-line call.

RECOMMENDATION: ACCEPT the substitution and close the epic.

Why the substitution is better than the design it replaced, not merely cheaper:
- A summarizer ROLE means a SECOND pi run per task whose only output is prose. That is another model call, another timeout to enforce, another attempt that can crash, stall or be killed — all on the finalize path, which is the one path that must not get more failure modes (see ah-1cx.2 for what a failure there already costs).
- The agent that did the work is the only party that knows WHY it made its choices. A summarizer reading the diff afterwards is reconstructing intent from artifacts; .task/summary.md is the agent stating it directly. The verdict-routing pipeline (ah-4el) already depends on that file's FIRST LINE, so the file is load-bearing regardless — a summarizer role would duplicate a channel that must exist anyway.
- mem0's infer=true already does the extraction step a summarizer would have done, on the server, at the point of storage.

What is actually lost: a run that ends WITHOUT writing summary.md contributes nothing to memory. Those are predominantly the failed, crashed and killed runs — exactly the ones a summarizer would have had least to say about, and where the failure comment already carries the diagnosis. If that gap ever matters, the cheap fix is a fallback that stores the diff stat + commit subjects, not a second agent.

If ACCEPTED: update this epic's body and SPEC §14.3's wording to describe the delivered design (SPEC §14 was already re-marked delivered in 119b84d and records the substitution as this epic's open decision — that sentence is what needs replacing), then close. That is a ~15 minute docs change and it is the last thing standing between v0.1.0 and its tag, alongside ah-1cx.8.

If REJECTED: the summarizer role is the one remaining piece of Stage 3, and it needs its own bead with a prompt, a role entry and a place on the finalize path.