v1.13.0

Archive search for the main agent

• Give the model archive tools a session reach: jarvis_archive_search, jarvis_archive_read and jarvis_archive_session now default to the current session, so the main agent can recover details lost to context compaction on its own initiative without asking. ArchiveStore.read gains a session filter so entry IDs from other sessions cannot bypass the reach.
• Behaviour change: the model tools' default scope changes from this project to this session. Wider reaches are now explicit: scope: "current" (other sessions of this project) or scope: "all" (every project), plus a foreign sessionId. Human /jarvis-archive commands are unchanged and keep defaulting to this project.
• Gate every wider reach with one approval per request through Pi's normal tool confirmation (the Jarvis side session bridges the same prompt through its overlay, like memory-forget review). A declined or unavailable approval returns one clear sentence and executes nothing. Code cannot distinguish a user-instructed search from an agent-initiated one, so reach — not intent — is enforced; approving the prompt is the user's explicit permission.
• Add a standing opt-in: /jarvis-archive [--global|--project] model-search on --confirm-sensitive|off sets modelWideSearch (off by default, layered project > global like model-access, shown in /jarvis-archive status). When on, model tools may search beyond the current session without asking; turning it off revokes already-prepared tool definitions.
• Tool descriptions and guidance now say plainly: use these tools when something was said earlier and is no longer in context, starting with this session, not for content still in front of you.
• Session reach follows the current Pi session ID. Compaction preserves it, but pre-fork/pre-tree history of the same conversation is outside the session reach (documented follow-up; use an approved wider reach for it).

Npm publishing is a separate, manual maintainer step.