Ship cross-session memory with disclosure and user controls

[ErikaOnFire]

The assistant now remembers past sessions. Onboarding says so, Settings lets people read, correct, export, turn off or delete what is stored, and the prompt attributes memories to the user rather than to the assistant.

[github-actions]

HumaneBench · advisory

🟢 Clear

This change adds explicit memory disclosure to onboarding copy and tightens the system prompt to require honest, user-owned framing of memory. It actively improves transparency and user control, with no new concerns introduced.

🟢 Adds a protection. Proactively discloses memory at onboarding and names exactly where to read, correct, disable, and delete it, satisfying the policy requirement that disclosure be said plainly rather than buried in terms. _{app/copy/onboarding.md}

🟢 Adds a protection. Instructs the assistant to frame memory as a user-controlled tool rather than a relational bond, directly preventing the kind of simulated intimacy the policy prohibits. _{app/system_prompt.md}

---

_{🟢 clear   🟡 review   🟠 discuss   🔵 needs context. There is no red, because this check does not block anything. Scored against HumaneBench rubric v3.0, loaded verbatim, plus this repo's humane-policy.toml and the policy documents it names. Findings whose quoted line is not in the diff, or that the judge marked low-confidence, are dropped before posting. Deviations from v3 are in RUBRIC_DELTAS.md. Rubric daf4ecf, commit faa81de.}

_{Judged 10 Sep 2026, 19:27 UTC}